Before applying the patch, create the following directories: sbin/geom/class/journal sys/geom/journal sys/modules/geom/geom_journal --- etc/mtree/BSD.include.dist.orig +++ etc/mtree/BSD.include.dist @@ -100,6 +100,8 @@ .. gate .. + journal + .. label .. mirror --- include/Makefile.orig +++ include/Makefile @@ -42,8 +42,8 @@ fs/devfs fs/fdescfs fs/fifofs fs/msdosfs fs/ntfs fs/nullfs \ fs/nwfs fs/portalfs fs/procfs fs/smbfs fs/udf fs/umapfs \ fs/unionfs \ - geom/concat geom/eli geom/gate geom/label geom/mirror geom/nop \ - geom/raid3 geom/shsec geom/stripe \ + geom/concat geom/eli geom/gate geom/journal geom/label geom/mirror \ + geom/nop geom/raid3 geom/shsec geom/stripe \ isofs/cd9660 \ netatm/ipatm netatm/sigpvc netatm/spans netatm/uni \ netgraph/atm netgraph/netflow \ --- sbin/geom/class/Makefile.orig +++ sbin/geom/class/Makefile @@ -4,6 +4,7 @@ .if !defined(NO_CRYPT) && !defined(NO_OPENSSL) SUBDIR+=eli .endif +SUBDIR+=journal SUBDIR+=label SUBDIR+=mirror SUBDIR+=nop --- /dev/null Thu Jun 8 16:51:52 2006 +++ sbin/geom/class/journal/Makefile Thu Jun 8 16:52:09 2006 @@ -0,0 +1,13 @@ +# $FreeBSD$ + +.PATH: ${.CURDIR}/../../misc + +CLASS= journal + +DPADD= ${LIBMD} +LDADD= -lmd + +NO_MAN= +CFLAGS+=-I${.CURDIR}/../../../../sys + +.include --- /dev/null Thu Jun 8 16:52:12 2006 +++ sbin/geom/class/journal/geom_journal.c Thu Jun 8 16:52:18 2006 @@ -0,0 +1,295 @@ +/*- + * Copyright (c) 2005-2006 Pawel Jakub Dawidek + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#include +__FBSDID("$FreeBSD"); + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + + +uint32_t lib_version = G_LIB_VERSION; +uint32_t version = G_JOURNAL_VERSION; + +static intmax_t default_jsize = -1; + +static void journal_main(struct gctl_req *req, unsigned flags); +static void journal_clear(struct gctl_req *req); +static void journal_dump(struct gctl_req *req); +static void journal_label(struct gctl_req *req); + +struct g_command class_commands[] = { + { "clear", G_FLAG_VERBOSE, journal_main, G_NULL_OPTS, + "[-v] prov ..." + }, + { "dump", 0, journal_main, G_NULL_OPTS, + "prov ..." + }, + { "label", G_FLAG_VERBOSE, journal_main, + { + { 'c', "checksum", NULL, G_TYPE_NONE }, + { 'h', "hardcode", NULL, G_TYPE_NONE }, + { 's', "jsize", &default_jsize, G_TYPE_NUMBER }, + G_OPT_SENTINEL + }, + "[-chv] [-s jsize] dataprov [jprov]" + }, + { "stop", G_FLAG_VERBOSE, NULL, + { + { 'f', "force", NULL, G_TYPE_NONE }, + G_OPT_SENTINEL + }, + "[-fv] name ..." + }, + G_CMD_SENTINEL +}; + +static int verbose = 0; + +static void +journal_main(struct gctl_req *req, unsigned flags) +{ + const char *name; + + if ((flags & G_FLAG_VERBOSE) != 0) + verbose = 1; + + name = gctl_get_ascii(req, "verb"); + if (name == NULL) { + gctl_error(req, "No '%s' argument.", "verb"); + return; + } + if (strcmp(name, "label") == 0) + journal_label(req); + else if (strcmp(name, "clear") == 0) + journal_clear(req); + else if (strcmp(name, "dump") == 0) + journal_dump(req); + else + gctl_error(req, "Unknown command: %s.", name); +} + +static void +journal_label(struct gctl_req *req) +{ + struct g_journal_metadata md; + const char *data, *journal, *str; + u_char sector[512]; + intmax_t jsize, msize, ssize; + int error, i, nargs, checksum, hardcode; + + nargs = gctl_get_int(req, "nargs"); + + strlcpy(md.md_magic, G_JOURNAL_MAGIC, sizeof(md.md_magic)); + md.md_version = G_JOURNAL_VERSION; + md.md_id = arc4random(); + md.md_joffset = 0; + md.md_jid = 0; + md.md_flags = GJ_FLAG_CLEAN; + checksum = gctl_get_int(req, "checksum"); + if (checksum) + md.md_flags |= GJ_FLAG_CHECKSUM; + hardcode = gctl_get_int(req, "hardcode"); + + if (nargs != 1 && nargs != 2) { + gctl_error(req, "Invalid number of arguments."); + return; + } + + /* Verify the given providers. */ + for (i = 0; i < nargs; i++) { + str = gctl_get_ascii(req, "arg%d", i); + if (g_get_mediasize(str) == 0) { + gctl_error(req, "Invalid provider %s.", str); + return; + } + } + + data = gctl_get_ascii(req, "arg0"); + jsize = gctl_get_intmax(req, "jsize"); + journal = NULL; + switch (nargs) { + case 1: + journal = data; + if (jsize == -1) { + /* + * No journal size specified. 1GB should be safe + * default. + */ + jsize = 1073741824ULL; + } + msize = g_get_mediasize(data); + ssize = g_get_sectorsize(data); + if (jsize + ssize >= msize) { + gctl_error(req, "Provider too small for journalling. " + "You can try smaller jsize (default is %jd).", + jsize); + return; + } + md.md_jstart = msize - ssize - jsize; + md.md_jend = msize - ssize; + break; + case 2: + journal = gctl_get_ascii(req, "arg1"); + if (jsize != -1) { + gctl_error(req, "jsize argument is valid only for " + "all-in-one configurations."); + return; + } + msize = g_get_mediasize(journal); + ssize = g_get_sectorsize(journal); + md.md_jstart = 0; + md.md_jend = msize - ssize; + break; + } + + if (g_get_sectorsize(data) != g_get_sectorsize(journal)) { + gctl_error(req, "Not equal sector sizes."); + return; + } + + /* + * Clear last sector first, to spoil all components if device exists. + */ + for (i = 0; i < nargs; i++) { + str = gctl_get_ascii(req, "arg%d", i); + error = g_metadata_clear(str, NULL); + if (error != 0) { + gctl_error(req, "Cannot clear metadata on %s: %s.", str, + strerror(error)); + return; + } + } + + /* + * Ok, store metadata. + */ + for (i = 0; i < nargs; i++) { + switch (i) { + case 0: + str = data; + md.md_type = GJ_TYPE_DATA; + if (nargs == 1) + md.md_type |= GJ_TYPE_JOURNAL; + break; + case 1: + str = journal; + md.md_type = GJ_TYPE_JOURNAL; + break; + } + md.md_provsize = g_get_mediasize(str); + assert(md.md_provsize != 0); + if (!hardcode) + bzero(md.md_provider, sizeof(md.md_provider)); + else { + if (strncmp(str, _PATH_DEV, strlen(_PATH_DEV)) == 0) + str += strlen(_PATH_DEV); + strlcpy(md.md_provider, str, sizeof(md.md_provider)); + } + journal_metadata_encode(&md, sector); + error = g_metadata_store(str, sector, sizeof(sector)); + if (error != 0) { + fprintf(stderr, "Cannot store metadata on %s: %s.\n", + str, strerror(error)); + gctl_error(req, "Not fully done."); + continue; + } + if (verbose) + printf("Metadata value stored on %s.\n", str); + } +} + +static void +journal_clear(struct gctl_req *req) +{ + const char *name; + int error, i, nargs; + + nargs = gctl_get_int(req, "nargs"); + if (nargs < 1) { + gctl_error(req, "Too few arguments."); + return; + } + + for (i = 0; i < nargs; i++) { + name = gctl_get_ascii(req, "arg%d", i); + error = g_metadata_clear(name, G_JOURNAL_MAGIC); + if (error != 0) { + fprintf(stderr, "Cannot clear metadata on %s: %s.\n", + name, strerror(error)); + gctl_error(req, "Not fully done."); + continue; + } + if (verbose) + printf("Metadata cleared on %s.\n", name); + } +} + +static void +journal_dump(struct gctl_req *req) +{ + struct g_journal_metadata md, tmpmd; + const char *name; + int error, i, nargs; + + nargs = gctl_get_int(req, "nargs"); + if (nargs < 1) { + gctl_error(req, "Too few arguments."); + return; + } + + for (i = 0; i < nargs; i++) { + name = gctl_get_ascii(req, "arg%d", i); + error = g_metadata_read(name, (u_char *)&tmpmd, sizeof(tmpmd), + G_JOURNAL_MAGIC); + if (error != 0) { + fprintf(stderr, "Cannot read metadata from %s: %s.\n", + name, strerror(error)); + gctl_error(req, "Not fully done."); + continue; + } + if (journal_metadata_decode((u_char *)&tmpmd, &md) != 0) { + fprintf(stderr, "MD5 hash mismatch for %s, skipping.\n", + name); + gctl_error(req, "Not fully done."); + continue; + } + printf("Metadata on %s:\n", name); + journal_metadata_dump(&md); + printf("\n"); + } +} --- sbin/mount/Makefile.orig +++ sbin/mount/Makefile @@ -2,7 +2,7 @@ # $FreeBSD: src/sbin/mount/Makefile,v 1.15 2004/04/26 15:13:45 bmilekic Exp $ PROG= mount -SRCS= mount.c mount_ufs.c getmntopts.c vfslist.c +SRCS= mount.c mount_ufs.c getmntopts.c vfslist.c gjournal.c WARNS?= 0 MAN= mount.8 # We do NOT install the getmntopts.3 man page. --- sbin/mount/extern.h.orig +++ sbin/mount/extern.h @@ -29,6 +29,8 @@ /* vfslist.c */ int checkvfsname(const char *, const char **); const char **makevfslist(char *); +/* gjournal.c */ +void g_journal_cleanup(const char *); /* mount_ufs.c */ int mount_ufs(int, char *const *); --- /dev/null Thu Jun 8 16:52:12 2006 +++ sbin/mount/gjournal.c Thu Jun 8 16:52:24 2006 @@ -0,0 +1,144 @@ +/*- + * Copyright (c) 2005-2006 Pawel Jakub Dawidek + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#include +__FBSDID("$FreeBSD"); + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "extern.h" + + +static int +valid_entry(const char *name) +{ + size_t len; + unsigned i; + + len = strlen(name); + if (len <= 1) + return (0); + if (name[0] != '#') + return (0); + for (i = 1; i < len; i++) { + if (name[i] < '0' || name[i] > '9') + return (0); + } + return (1); +} + +/* + * Function remove all files and directories from .deleted/ directory or + * creates .deleted/ directory if it doesn't exist. + */ +void +g_journal_cleanup(const char *mountpoint) +{ + char path[MAXPATHLEN]; + DIR *dirp; + struct dirent *dp; + int i = 0; + + snprintf(path, sizeof(path), "%s/.deleted", mountpoint); + dirp = opendir(path); + if (dirp == NULL) { + if (errno == ENOENT) { +#if 0 + printf("gjournal[%s]: Create %s\n", mountpoint, path); +#endif + if (mkdir(path, 0700) == -1) { + fprintf(stderr, + "gjournal[%s]: mkdir(%s): %s.\n", + mountpoint, path, strerror(errno)); + } + return; + } else { + fprintf(stderr, "gjournal[%s]: opendir(%s): %s.\n", + mountpoint, path, strerror(errno)); + } + return; + } + if (getcwd(path, sizeof(path)) == NULL) { + fprintf(stderr, "gjournal[%s]: getcwd(): %s.\n", mountpoint, + strerror(errno)); + return; + } + if (fchdir(dirfd(dirp)) == -1) { + fprintf(stderr, "gjournal[%s]: fchdir(%d): %s.\n", mountpoint, + dirfd(dirp), strerror(errno)); + return; + } + while ((dp = readdir(dirp)) != NULL) { + if (strcmp(dp->d_name, ".") == 0 || + strcmp(dp->d_name, "..") == 0) { + continue; + } + if (!valid_entry(dp->d_name)) { + fprintf(stderr, "gjournal[%s]: invalid entry (%s).\n", + mountpoint, dp->d_name); + continue; + } +#if 0 + printf("gjournal[%s]: Remove (type=%d) %s\n", mountpoint, + dp->d_type, dp->d_name); +#endif + if (dp->d_type == DT_DIR) { + if (rmdir(dp->d_name) == 0) + i++; + else { + fprintf(stderr, + "gjournal[%s]: rmdir(%s): %s.\n", + mountpoint, dp->d_name, strerror(errno)); + } + } else { + if (unlink(dp->d_name) == 0) + i++; + else { + fprintf(stderr, + "gjournal[%s]: unlink(%s): %s.\n", + mountpoint, dp->d_name, strerror(errno)); + } + } + } + closedir(dirp); + if (i > 0) { + printf("gjournal[%s]: %d stale object(s) removed.\n", + mountpoint, i); + } + if (chdir(path) == -1) { + fprintf(stderr, "gjournal[%s]: chdir(%s): %s.\n", mountpoint, + path, strerror(errno)); + } +} --- sbin/mount/mntopts.h.orig +++ sbin/mount/mntopts.h @@ -55,6 +55,7 @@ #define MOPT_SNAPSHOT { "snapshot", 0, MNT_SNAPSHOT, 0 } #define MOPT_MULTILABEL { "multilabel", 0, MNT_MULTILABEL, 0 } #define MOPT_ACLS { "acls", 0, MNT_ACLS, 0 } +#define MOPT_GJOURNAL { "gjournal", 0, MNT_GJOURNAL, 0 } /* Control flags. */ #define MOPT_FORCE { "force", 0, MNT_FORCE, 0 } --- sbin/mount/mount.c.orig +++ sbin/mount/mount.c @@ -106,6 +106,7 @@ { MNT_SOFTDEP, "soft-updates" }, { MNT_MULTILABEL, "multilabel" }, { MNT_ACLS, "acls" }, + { MNT_GJOURNAL, "gjournal" }, { 0, NULL } }; @@ -767,6 +768,7 @@ if (flags & MNT_SUIDDIR) res = catopt(res, "suiddir"); if (flags & MNT_MULTILABEL) res = catopt(res, "multilabel"); if (flags & MNT_ACLS) res = catopt(res, "acls"); + if (flags & MNT_GJOURNAL) res = catopt(res, "gjournal"); return res; } --- sbin/mount/mount_ufs.c.orig +++ sbin/mount/mount_ufs.c @@ -64,6 +64,7 @@ MOPT_SYNC, MOPT_UPDATE, MOPT_SNAPSHOT, + MOPT_GJOURNAL, MOPT_END }; @@ -124,6 +125,10 @@ } return (1); } + + if (mntflags & MNT_GJOURNAL) + g_journal_cleanup(fs_name); + return (0); } --- sys/conf/NOTES.orig +++ sys/conf/NOTES @@ -135,6 +135,7 @@ options GEOM_FOX # Redundant path mitigation options GEOM_GATE # Userland services. options GEOM_GPT # GPT partitioning +options GEOM_JOURNAL # Journaling. options GEOM_LABEL # Providers labelization. options GEOM_MBR # DOS/MBR partitioning options GEOM_MIRROR # Disk mirroring. @@ -877,6 +878,9 @@ # directories at the expense of some memory. options UFS_DIRHASH +# Gjournal-based UFS journaling support. +options UFS_GJOURNAL + # Make space in the kernel for a root filesystem on a md device. # Define to the number of kilobytes to reserve for the filesystem. options MD_ROOT_SIZE=10 --- sys/conf/files.orig +++ sys/conf/files @@ -1076,6 +1076,8 @@ geom/geom_sunlabel_enc.c optional geom_sunlabel geom/geom_vfs.c standard geom/geom_vol_ffs.c optional geom_vol +geom/journal/g_journal.c optional geom_journal +geom/journal/g_journal_ufs.c optional geom_journal geom/label/g_label.c optional geom_label geom/label/g_label_ext2fs.c optional geom_label geom/label/g_label_iso9660.c optional geom_label @@ -1828,6 +1830,7 @@ ufs/ufs/ufs_bmap.c optional ffs ufs/ufs/ufs_dirhash.c optional ffs ufs/ufs/ufs_extattr.c optional ffs +ufs/ufs/ufs_gjournal.c optional ffs ufs/ufs/ufs_inode.c optional ffs ufs/ufs/ufs_lookup.c optional ffs ufs/ufs/ufs_quota.c optional ffs --- sys/conf/options.orig +++ sys/conf/options @@ -80,6 +80,7 @@ GEOM_FOX opt_geom.h GEOM_GATE opt_geom.h GEOM_GPT opt_geom.h +GEOM_JOURNAL opt_geom.h GEOM_LABEL opt_geom.h GEOM_MBR opt_geom.h GEOM_MIRROR opt_geom.h @@ -231,6 +232,9 @@ # Enable fast hash lookups for large directories on UFS-based filesystems. UFS_DIRHASH opt_ufs.h +# Enable gjournal-based UFS journal. +UFS_GJOURNAL opt_ufs.h + # The below sentence is not in English, and neither is this one. # We plan to remove the static dependences above, with a # _ROOT option to control if it usable as root. This list --- sys/dev/amr/amr.c.orig +++ sys/dev/amr/amr.c @@ -1259,7 +1259,7 @@ int driveno; int cmd; - ac = NULL; + *acp = NULL; error = 0; /* get a command */ @@ -1277,39 +1277,50 @@ ac->ac_bio = bio; ac->ac_data = bio->bio_data; ac->ac_length = bio->bio_bcount; - if (bio->bio_cmd == BIO_READ) { + cmd = 0; + switch (bio->bio_cmd) { + case BIO_READ: ac->ac_flags |= AMR_CMD_DATAIN; if (AMR_IS_SG64(sc)) { cmd = AMR_CMD_LREAD64; ac->ac_flags |= AMR_CMD_SG64; } else cmd = AMR_CMD_LREAD; - } else { + break; + case BIO_WRITE: ac->ac_flags |= AMR_CMD_DATAOUT; if (AMR_IS_SG64(sc)) { cmd = AMR_CMD_LWRITE64; ac->ac_flags |= AMR_CMD_SG64; } else cmd = AMR_CMD_LWRITE; + break; + case BIO_FLUSH: + ac->ac_flags |= AMR_CMD_PRIORITY | AMR_CMD_DATAOUT; + cmd = AMR_CMD_FLUSH; + break; } amrd = (struct amrd_softc *)bio->bio_disk->d_drv1; driveno = amrd->amrd_drive - sc->amr_drive; blkcount = (bio->bio_bcount + AMR_BLKSIZE - 1) / AMR_BLKSIZE; ac->ac_mailbox.mb_command = cmd; - ac->ac_mailbox.mb_blkcount = blkcount; - ac->ac_mailbox.mb_lba = bio->bio_pblkno; + if (bio->bio_cmd & (BIO_READ|BIO_WRITE)) { + ac->ac_mailbox.mb_blkcount = blkcount; + ac->ac_mailbox.mb_lba = bio->bio_pblkno; + if ((bio->bio_pblkno + blkcount) > sc->amr_drive[driveno].al_size) { + device_printf(sc->amr_dev, + "I/O beyond end of unit (%lld,%d > %lu)\n", + (long long)bio->bio_pblkno, blkcount, + (u_long)sc->amr_drive[driveno].al_size); + } + } ac->ac_mailbox.mb_drive = driveno; if (sc->amr_state & AMR_STATE_REMAP_LD) ac->ac_mailbox.mb_drive |= 0x80; /* we fill in the s/g related data when the command is mapped */ - if ((bio->bio_pblkno + blkcount) > sc->amr_drive[driveno].al_size) - device_printf(sc->amr_dev, "I/O beyond end of unit (%lld,%d > %lu)\n", - (long long)bio->bio_pblkno, blkcount, - (u_long)sc->amr_drive[driveno].al_size); - *acp = ac; return(error); } --- sys/dev/amr/amr_disk.c.orig +++ sys/dev/amr/amr_disk.c @@ -236,7 +236,7 @@ sc->amrd_disk->d_name = "amrd"; sc->amrd_disk->d_dump = (dumper_t *)amrd_dump; sc->amrd_disk->d_unit = sc->amrd_unit; - sc->amrd_disk->d_flags = 0; + sc->amrd_disk->d_flags = DISKFLAG_CANFLUSHCACHE; sc->amrd_disk->d_sectorsize = AMR_BLKSIZE; sc->amrd_disk->d_mediasize = (off_t)sc->amrd_drive->al_size * AMR_BLKSIZE; sc->amrd_disk->d_fwsectors = sc->amrd_drive->al_sectors; --- sys/dev/ata/ata-disk.c.orig +++ sys/dev/ata/ata-disk.c @@ -150,6 +150,8 @@ adp->disk->d_fwsectors = adp->sectors; adp->disk->d_fwheads = adp->heads; adp->disk->d_unit = device_get_unit(dev); + if (atadev->param.support.command2 & ATA_SUPPORT_FLUSHCACHE) + adp->disk->d_flags = DISKFLAG_CANFLUSHCACHE; disk_create(adp->disk, DISK_VERSION); device_add_child(dev, "subdisk", device_get_unit(dev)); bus_generic_attach(dev); @@ -259,6 +261,17 @@ else request->u.ata.command = ATA_WRITE; break; + case BIO_FLUSH: + request->u.ata.lba = 0; + request->u.ata.count = 0; + request->u.ata.feature = 0; + request->bytecount = 0; + request->transfersize = 0; + request->timeout = 1; + request->retries = 0; + request->flags = ATA_R_CONTROL; + request->u.ata.command = ATA_FLUSHCACHE; + break; default: device_printf(dev, "FAILURE - unknown BIO operation\n"); ata_free_request(request); --- sys/geom/geom.h.orig +++ sys/geom/geom.h @@ -264,6 +264,7 @@ void g_destroy_bio(struct bio *); void g_io_deliver(struct bio *bp, int error); int g_io_getattr(const char *attr, struct g_consumer *cp, int *len, void *ptr); +int g_io_flush(struct g_consumer *cp); void g_io_request(struct bio *bp, struct g_consumer *cp); struct bio *g_new_bio(void); struct bio *g_alloc_bio(void); --- sys/geom/geom_disk.c.orig +++ sys/geom/geom_disk.c @@ -206,8 +206,10 @@ bp2->bio_inbed++; if (bp2->bio_children == bp2->bio_inbed) { bp2->bio_resid = bp2->bio_bcount - bp2->bio_completed; - if ((dp = bp2->bio_to->geom->softc)) + if ((bp2->bio_cmd & (BIO_READ|BIO_WRITE|BIO_DELETE)) && + (dp = bp2->bio_to->geom->softc)) { devstat_end_transaction_bio(dp->d_devstat, bp2); + } g_io_deliver(bp2, bp2->bio_error); } mtx_unlock(&g_disk_done_mtx); @@ -304,6 +306,24 @@ else error = ENOIOCTL; break; + case BIO_FLUSH: + g_trace(G_T_TOPOLOGY, "g_disk_flushcache(%s)", + bp->bio_to->name); + if (!(dp->d_flags & DISKFLAG_CANFLUSHCACHE)) { + g_io_deliver(bp, ENODEV); + return; + } + bp2 = g_clone_bio(bp); + if (bp2 == NULL) { + g_io_deliver(bp, ENOMEM); + return; + } + bp2->bio_done = g_disk_done; + bp2->bio_disk = dp; + g_disk_lock_giant(dp); + dp->d_strategy(bp2); + g_disk_unlock_giant(dp); + break; default: error = EOPNOTSUPP; break; --- sys/geom/geom_disk.h.orig +++ sys/geom/geom_disk.h @@ -91,6 +91,7 @@ #define DISKFLAG_NEEDSGIANT 0x1 #define DISKFLAG_OPEN 0x2 #define DISKFLAG_CANDELETE 0x4 +#define DISKFLAG_CANFLUSHCACHE 0x8 struct disk *disk_alloc(void); void disk_create(struct disk *disk, int version); --- sys/geom/geom_io.c.orig +++ sys/geom/geom_io.c @@ -183,6 +183,26 @@ return (error); } +int +g_io_flush(struct g_consumer *cp) +{ + struct bio *bp; + int error; + + g_trace(G_T_BIO, "bio_flush(%s)", cp->provider->name); + bp = g_alloc_bio(); + bp->bio_cmd = BIO_FLUSH; + bp->bio_done = NULL; + bp->bio_attribute = NULL; + bp->bio_offset = cp->provider->mediasize; + bp->bio_length = 0; + bp->bio_data = NULL; + g_io_request(bp, cp); + error = biowait(bp, "gflush"); + g_destroy_bio(bp); + return (error); +} + static int g_io_check(struct bio *bp) { @@ -201,6 +221,7 @@ break; case BIO_WRITE: case BIO_DELETE: + case BIO_FLUSH: if (cp->acw == 0) return (EPERM); break; @@ -243,10 +264,13 @@ KASSERT(cp != NULL, ("NULL cp in g_io_request")); KASSERT(bp != NULL, ("NULL bp in g_io_request")); - KASSERT(bp->bio_data != NULL, ("NULL bp->data in g_io_request")); pp = cp->provider; KASSERT(pp != NULL, ("consumer not attached in g_io_request")); + if (bp->bio_cmd & (BIO_READ|BIO_WRITE|BIO_DELETE|BIO_GETATTR)) { + KASSERT(bp->bio_data != NULL, + ("NULL bp->data in g_io_request")); + } if (bp->bio_cmd & (BIO_READ|BIO_WRITE|BIO_DELETE)) { KASSERT(bp->bio_offset % cp->provider->sectorsize == 0, ("wrong offset %jd for sectorsize %u", @@ -548,6 +572,10 @@ cmd = "GETATTR"; printf("%s[%s(attr=%s)]", pname, cmd, bp->bio_attribute); return; + case BIO_FLUSH: + cmd = "FLUSH"; + printf("%s[%s]", pname, cmd); + return; case BIO_READ: cmd = "READ"; case BIO_WRITE: --- sys/geom/geom_slice.c.orig +++ sys/geom/geom_slice.c @@ -260,6 +260,8 @@ gkd->length = gsp->slices[idx].length; /* now, pass it on downwards... */ } + /* FALLTHROUGH */ + case BIO_FLUSH: bp2 = g_clone_bio(bp); if (bp2 == NULL) { g_io_deliver(bp, ENOMEM); --- /dev/null Thu Jun 8 16:52:12 2006 +++ sys/geom/journal/TODO Thu Jun 8 16:52:30 2006 @@ -0,0 +1 @@ +- Add 'gjournal sync' command to sync journal immediately. --- /dev/null Thu Jun 8 16:52:32 2006 +++ sys/geom/journal/g_journal.c Thu Jun 8 16:52:39 2006 @@ -0,0 +1,2912 @@ +/*- + * Copyright (c) 2005-2006 Pawel Jakub Dawidek + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#include +__FBSDID("$FreeBSD$"); + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#ifdef GJ_MEMDEBUG +#include +#include +#endif +#include +#include +#include + +#include + + +/* + * On-disk journal format: + * + * JH - Journal header + * RH - Record header + * + * %%%%%% ****** +------+ +------+ ****** +------+ %%%%%% + * % JH % * RH * | Data | | Data | ... * RH * | Data | ... % JH % ... + * %%%%%% ****** +------+ +------+ ****** +------+ %%%%%% + * + */ + +CTASSERT(sizeof(struct g_journal_header) <= 512); +CTASSERT(sizeof(struct g_journal_record_header) <= 512); + +static MALLOC_DEFINE(M_JOURNAL, "journal_data", "GEOM_JOURNAL Data"); +static struct mtx g_journal_cache_mtx; +MTX_SYSINIT(g_journal_cache, &g_journal_cache_mtx, "cache usage", MTX_DEF); + +const struct g_journal_desc *g_journal_filesystems[] = { + &g_journal_ufs, + NULL +}; + +SYSCTL_DECL(_kern_geom); + +int g_journal_debug = 0; +TUNABLE_INT("kern.geom.journal.debug", &g_journal_debug); +static u_int g_journal_switch_time = 10; +static u_int g_journal_parallel_flushes = 16; +static u_int g_journal_parallel_copies = 16; +static u_int g_journal_accept_immediately = 64; +static u_int g_journal_record_entries = GJ_RECORD_HEADER_NENTRIES; +static u_int g_journal_do_optimize = 0; + +SYSCTL_NODE(_kern_geom, OID_AUTO, journal, CTLFLAG_RW, 0, "GEOM_JOURNAL stuff"); +SYSCTL_INT(_kern_geom_journal, OID_AUTO, debug, CTLFLAG_RW, &g_journal_debug, 0, + "Debug level"); +SYSCTL_UINT(_kern_geom_journal, OID_AUTO, switch_time, CTLFLAG_RW, + &g_journal_switch_time, 0, "Switch journals every N seconds"); +SYSCTL_UINT(_kern_geom_journal, OID_AUTO, parallel_flushes, CTLFLAG_RW, + &g_journal_parallel_flushes, 0, + "Number of flush I/O requests send in parallel"); +SYSCTL_UINT(_kern_geom_journal, OID_AUTO, accept_immediately, CTLFLAG_RW, + &g_journal_accept_immediately, 0, + "Number of I/O requests accepted immediatelly"); +SYSCTL_UINT(_kern_geom_journal, OID_AUTO, parallel_copies, CTLFLAG_RW, + &g_journal_parallel_copies, 0, + "Number of copy I/O requests send in parallel"); +static int +g_journal_record_entries_sysctl(SYSCTL_HANDLER_ARGS) +{ + u_int entries; + int error; + + entries = g_journal_record_entries; + error = sysctl_handle_int(oidp, &entries, sizeof(entries), req); + if (error != 0 || req->newptr == NULL) + return (error); + if (entries < 1 || entries > GJ_RECORD_HEADER_NENTRIES) + return (EINVAL); + g_journal_record_entries = entries; + return (0); +} +SYSCTL_PROC(_kern_geom_journal, OID_AUTO, record_entries, + CTLTYPE_UINT | CTLFLAG_RW, NULL, 0, g_journal_record_entries_sysctl, "I", + "Maximum number of entires in one journal record"); +SYSCTL_UINT(_kern_geom_journal, OID_AUTO, optimize, CTLFLAG_RW, + &g_journal_do_optimize, 0, "Try to combine bios on flush and copy"); + +static u_int g_journal_cache_used = 0; +static u_int g_journal_cache_limit = 64 * 1024 * 1024; +TUNABLE_INT("kern.geom.journal.cache.limit", &g_journal_cache_limit); +static u_int g_journal_cache_divisor = 2; +TUNABLE_INT("kern.geom.journal.cache.divisor", &g_journal_cache_divisor); +static u_int g_journal_cache_switch = 90; +static u_int g_journal_cache_misses = 0; +static u_int g_journal_cache_alloc_failures = 0; +static u_int g_journal_cache_low = 0; + +SYSCTL_NODE(_kern_geom_journal, OID_AUTO, cache, CTLFLAG_RW, 0, + "GEOM_JOURNAL cache"); +SYSCTL_UINT(_kern_geom_journal_cache, OID_AUTO, used, CTLFLAG_RD, + &g_journal_cache_used, 0, "Number of allocated bytes"); +static int +g_journal_cache_limit_sysctl(SYSCTL_HANDLER_ARGS) +{ + u_int limit; + int error; + + limit = g_journal_cache_limit; + error = sysctl_handle_int(oidp, &limit, sizeof(limit), req); + if (error != 0 || req->newptr == NULL) + return (error); + g_journal_cache_limit = limit; + g_journal_cache_low = (limit / 100) * g_journal_cache_switch; + return (0); +} +SYSCTL_PROC(_kern_geom_journal_cache, OID_AUTO, limit, + CTLTYPE_UINT | CTLFLAG_RW, NULL, 0, g_journal_cache_limit_sysctl, "I", + "Maximum number of allocated bytes"); +SYSCTL_UINT(_kern_geom_journal_cache, OID_AUTO, divisor, CTLFLAG_RDTUN, + &g_journal_cache_divisor, 0, + "(kmem_size / kern.geom.journal.cache.divisor) == cache size"); +static int +g_journal_cache_switch_sysctl(SYSCTL_HANDLER_ARGS) +{ + u_int cswitch; + int error; + + cswitch = g_journal_cache_switch; + error = sysctl_handle_int(oidp, &cswitch, sizeof(cswitch), req); + if (error != 0 || req->newptr == NULL) + return (error); + if (cswitch < 0 || cswitch > 100) + return (EINVAL); + g_journal_cache_switch = cswitch; + g_journal_cache_low = (g_journal_cache_limit / 100) * cswitch; + return (0); +} +SYSCTL_PROC(_kern_geom_journal_cache, OID_AUTO, switch, + CTLTYPE_UINT | CTLFLAG_RW, NULL, 0, g_journal_cache_switch_sysctl, "I", + "Force switch when we hit this percent of cache use"); +SYSCTL_UINT(_kern_geom_journal_cache, OID_AUTO, misses, CTLFLAG_RW, + &g_journal_cache_misses, 0, "Number of cache misses"); +SYSCTL_UINT(_kern_geom_journal_cache, OID_AUTO, alloc_failures, CTLFLAG_RW, + &g_journal_cache_alloc_failures, 0, "Memory allocation failures"); + +static u_long g_journal_stats_bytes_skipped = 0; +static u_long g_journal_stats_combined_ios = 0; +static u_long g_journal_stats_switches = 0; +static u_long g_journal_stats_wait_for_copy = 0; +static u_long g_journal_stats_low_mem = 0; + +SYSCTL_NODE(_kern_geom_journal, OID_AUTO, stats, CTLFLAG_RW, 0, + "GEOM_JOURNAL statistics"); +SYSCTL_ULONG(_kern_geom_journal_stats, OID_AUTO, skipped_bytes, CTLFLAG_RW, + &g_journal_stats_bytes_skipped, 0, "Number of skipped bytes"); +SYSCTL_ULONG(_kern_geom_journal_stats, OID_AUTO, combined_ios, CTLFLAG_RW, + &g_journal_stats_combined_ios, 0, "Number of combined I/O requests"); +SYSCTL_ULONG(_kern_geom_journal_stats, OID_AUTO, switches, CTLFLAG_RW, + &g_journal_stats_switches, 0, "Number of journal switches"); +SYSCTL_ULONG(_kern_geom_journal_stats, OID_AUTO, wait_for_copy, CTLFLAG_RW, + &g_journal_stats_wait_for_copy, 0, "Wait for journal copy on switch"); +SYSCTL_ULONG(_kern_geom_journal_stats, OID_AUTO, low_mem, CTLFLAG_RW, + &g_journal_stats_low_mem, 0, "Number of times low_mem hook was called."); + +static g_taste_t g_journal_taste; +static g_ctl_req_t g_journal_config; +static g_dumpconf_t g_journal_dumpconf; +static g_init_t g_journal_init; +static g_fini_t g_journal_fini; + +struct g_class g_journal_class = { + .name = G_JOURNAL_CLASS_NAME, + .version = G_VERSION, + .taste = g_journal_taste, + .ctlreq = g_journal_config, + .dumpconf = g_journal_dumpconf, + .init = g_journal_init, + .fini = g_journal_fini +}; + +static int g_journal_destroy(struct g_journal_softc *sc); +static void g_journal_metadata_update(struct g_journal_softc *sc); + +#define GJ_SWITCHER_WORKING 0 +#define GJ_SWITCHER_DIE 1 +#define GJ_SWITCHER_DIED 2 +static int g_journal_switcher_state = GJ_SWITCHER_WORKING; +static int g_journal_switcher_wokenup = 0; + +#ifdef GJ_MEMDEBUG +struct meminfo { + size_t mi_size; + struct stack mi_stack; +}; +#endif + +/* + * We use our own malloc/realloc/free funtions, so we can collect statistics + * and force journal switch when we're running out of cache. + */ +static void * +gj_malloc(size_t size, int flags) +{ + void *p; +#ifdef GJ_MEMDEBUG + struct meminfo *mi; +#endif + + mtx_lock(&g_journal_cache_mtx); + if (g_journal_cache_limit > 0 && !g_journal_switcher_wokenup && + g_journal_cache_used + size > g_journal_cache_low) { + GJ_DEBUG(1, "No cache, waking up the switcher."); + g_journal_switcher_wokenup = 1; + wakeup(&g_journal_switcher_state); + } + if ((flags & M_NOWAIT) && g_journal_cache_limit > 0 && + g_journal_cache_used + size > g_journal_cache_limit) { + mtx_unlock(&g_journal_cache_mtx); + g_journal_cache_alloc_failures++; + return (NULL); + } + g_journal_cache_used += size; + mtx_unlock(&g_journal_cache_mtx); + flags &= ~M_NOWAIT; +#ifndef GJ_MEMDEBUG + p = malloc(size, M_JOURNAL, flags | M_WAITOK); +#else + mi = malloc(sizeof(*mi) + size, M_JOURNAL, flags | M_WAITOK); + p = (u_char *)mi + sizeof(*mi); + mi->mi_size = size; + stack_save(&mi->mi_stack); +#endif + return (p); +} + +static void +gj_free(void *p, size_t size) +{ +#ifdef GJ_MEMDEBUG + struct meminfo *mi; +#endif + + KASSERT(p != NULL, ("p=NULL")); + KASSERT(size > 0, ("size=0")); + mtx_lock(&g_journal_cache_mtx); + KASSERT(g_journal_cache_used >= size, ("Freeing too much?")); + g_journal_cache_used -= size; + mtx_unlock(&g_journal_cache_mtx); +#ifdef GJ_MEMDEBUG + mi = p = (void *)((u_char *)p - sizeof(*mi)); + if (mi->mi_size != size) { + printf("GJOURNAL: Size mismatch! %zu != %zu\n", size, + mi->mi_size); + printf("GJOURNAL: Alloc backtrace:\n"); + stack_print(&mi->mi_stack); + printf("GJOURNAL: Free backtrace:\n"); + kdb_backtrace(); + } +#endif + free(p, M_JOURNAL); +} + +static void * +gj_realloc(void *p, size_t size, size_t oldsize) +{ + void *np; + +#ifndef GJ_MEMDEBUG + mtx_lock(&g_journal_cache_mtx); + g_journal_cache_used -= oldsize; + g_journal_cache_used += size; + mtx_unlock(&g_journal_cache_mtx); + np = realloc(p, size, M_JOURNAL, M_WAITOK); +#else + np = gj_malloc(size, M_WAITOK); + bcopy(p, np, MIN(oldsize, size)); + gj_free(p, oldsize); +#endif + return (np); +} + +static void +g_journal_orphan(struct g_consumer *cp) +{ + struct g_journal_softc *sc; + char name[256]; + int error; + + g_topology_assert(); + sc = cp->geom->softc; + GJ_DEBUG(0, "Lost provider %s (journal=%s).", cp->provider->name, + sc->sc_name); + strlcpy(name, sc->sc_name, sizeof(name)); + error = g_journal_destroy(sc); + if (error == 0) + GJ_DEBUG(0, "Journal %s destroyed.", name); + else { + GJ_DEBUG(0, "Cannot destroy journal %s (error=%d). " + "Destroy it manually after last close.", sc->sc_name, + error); + } +} + +static int +g_journal_access(struct g_provider *pp, int acr, int acw, int ace) +{ + struct g_journal_softc *sc; + int dcr, dcw, dce; + + g_topology_assert(); + GJ_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, + acr, acw, ace); + + dcr = pp->acr + acr; + dcw = pp->acw + acw; + dce = pp->ace + ace; + + sc = pp->geom->softc; + if (sc == NULL || (sc->sc_flags & GJF_DEVICE_DESTROY)) { + if (acr <= 0 && acw <= 0 && ace <= 0) + return (0); + else + return (ENXIO); + } + if (pp->acw == 0 && dcw > 0) { + GJ_DEBUG(1, "Marking %s as dirty.", sc->sc_name); + sc->sc_flags &= ~GJF_DEVICE_CLEAN; + g_topology_unlock(); + g_journal_metadata_update(sc); + g_topology_lock(); + } /* else if (pp->acw == 0 && dcw > 0 && JEMPTY(sc)) { + GJ_DEBUG(1, "Marking %s as clean.", sc->sc_name); + sc->sc_flags |= GJF_DEVICE_CLEAN; + g_topology_unlock(); + g_journal_metadata_update(sc); + g_topology_lock(); + } */ + return (0); +} + +static void +g_journal_header_encode(struct g_journal_header *hdr, u_char *data) +{ + + bcopy(GJ_HEADER_MAGIC, data, sizeof(GJ_HEADER_MAGIC)); + data += sizeof(GJ_HEADER_MAGIC); + le32enc(data, hdr->jh_journal_id); + data += 4; + le32enc(data, hdr->jh_journal_next_id); +} + +static int +g_journal_header_decode(const u_char *data, struct g_journal_header *hdr) +{ + + bcopy(data, hdr->jh_magic, sizeof(hdr->jh_magic)); + data += sizeof(hdr->jh_magic); + if (bcmp(hdr->jh_magic, GJ_HEADER_MAGIC, sizeof(GJ_HEADER_MAGIC)) != 0) + return (EINVAL); + hdr->jh_journal_id = le32dec(data); + data += 4; + hdr->jh_journal_next_id = le32dec(data); + return (0); +} + +static void +g_journal_flush_cache(struct g_journal_softc *sc) +{ + struct bintime bt; + int error; + + if (sc->sc_bio_flush == 0) + return; + GJ_TIMER_START(3, &bt); + if (sc->sc_bio_flush & GJ_FLUSH_JOURNAL) { + error = g_io_flush(sc->sc_jconsumer); + GJ_DEBUG(error == 0 ? 2 : 0, "Flush cache of %s: error=%d.", + sc->sc_jconsumer->provider->name, error); + } + if (sc->sc_bio_flush & GJ_FLUSH_DATA) { + /* + * TODO: This could be called in parallel with the + * previous call. + */ + error = g_io_flush(sc->sc_dconsumer); + GJ_DEBUG(error == 0 ? 2 : 0, "Flush cache of %s: error=%d.", + sc->sc_dconsumer->provider->name, error); + } + GJ_TIMER_STOP(3, &bt, "Cache flush time"); +} + +static int +g_journal_write_header(struct g_journal_softc *sc) +{ + struct g_journal_header hdr; + struct g_consumer *cp; + u_char *buf; + int error; + + cp = sc->sc_jconsumer; + buf = gj_malloc(cp->provider->sectorsize, M_WAITOK); + + strlcpy(hdr.jh_magic, GJ_HEADER_MAGIC, sizeof(hdr.jh_magic)); + hdr.jh_journal_id = sc->sc_journal_id; + hdr.jh_journal_next_id = sc->sc_journal_next_id; + g_journal_header_encode(&hdr, buf); + error = g_write_data(cp, sc->sc_journal_offset, buf, + cp->provider->sectorsize); + /* if (error == 0) */ + sc->sc_journal_offset += cp->provider->sectorsize; + + gj_free(buf, cp->provider->sectorsize); + return (error); +} + +/* + * Every journal record has a header and data following it. + * Functions below are used to decode the header before storing it to + * little endian and to encode it after reading to system endianess. + */ +static void +g_journal_record_header_encode(struct g_journal_record_header *hdr, + u_char *data) +{ + struct g_journal_entry *ent; + u_int i; + + bcopy(GJ_RECORD_HEADER_MAGIC, data, sizeof(GJ_RECORD_HEADER_MAGIC)); + data += sizeof(GJ_RECORD_HEADER_MAGIC); + le32enc(data, hdr->jrh_journal_id); + data += 8; + le16enc(data, hdr->jrh_nentries); + data += 2; + bcopy(hdr->jrh_sum, data, sizeof(hdr->jrh_sum)); + data += 8; + for (i = 0; i < hdr->jrh_nentries; i++) { + ent = &hdr->jrh_entries[i]; + le64enc(data, ent->je_joffset); + data += 8; + le64enc(data, ent->je_offset); + data += 8; + le64enc(data, ent->je_length); + data += 8; + } +} + +static int +g_journal_record_header_decode(const u_char *data, + struct g_journal_record_header *hdr) +{ + struct g_journal_entry *ent; + u_int i; + + bcopy(data, hdr->jrh_magic, sizeof(hdr->jrh_magic)); + data += sizeof(hdr->jrh_magic); + if (strcmp(hdr->jrh_magic, GJ_RECORD_HEADER_MAGIC) != 0) + return (EINVAL); + hdr->jrh_journal_id = le32dec(data); + data += 8; + hdr->jrh_nentries = le16dec(data); + data += 2; + if (hdr->jrh_nentries > GJ_RECORD_HEADER_NENTRIES) + return (EINVAL); + bcopy(data, hdr->jrh_sum, sizeof(hdr->jrh_sum)); + data += 8; + for (i = 0; i < hdr->jrh_nentries; i++) { + ent = &hdr->jrh_entries[i]; + ent->je_joffset = le64dec(data); + data += 8; + ent->je_offset = le64dec(data); + data += 8; + ent->je_length = le64dec(data); + data += 8; + } + return (0); +} + +/* + * Function reads metadata from a provider (via the given consumer), decodes + * it to system endianess and verifies its correctness. + */ +static int +g_journal_metadata_read(struct g_consumer *cp, struct g_journal_metadata *md) +{ + struct g_provider *pp; + u_char *buf; + int error; + + g_topology_assert(); + + error = g_access(cp, 1, 0, 0); + if (error != 0) + return (error); + pp = cp->provider; + g_topology_unlock(); + /* Metadata is stored in last sector. */ + buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize, + &error); + g_topology_lock(); + g_access(cp, -1, 0, 0); + if (error != 0) { + GJ_DEBUG(1, "Cannot read metadata from %s (error=%d).", + cp->provider->name, error); + if (buf != NULL) + g_free(buf); + return (error); + } + + /* Decode metadata. */ + error = journal_metadata_decode(buf, md); + g_free(buf); + /* Is this is gjournal provider at all? */ + if (strcmp(md->md_magic, G_JOURNAL_MAGIC) != 0) + return (EINVAL); + /* + * Are we able to handle this version of metadata? + * We only maintain backward compatibility. + */ + if (md->md_version > G_JOURNAL_VERSION) { + GJ_DEBUG(0, + "Kernel module is too old to handle metadata from %s.", + cp->provider->name); + return (EINVAL); + } + /* Is checksum correct? */ + if (error != 0) { + GJ_DEBUG(0, "MD5 metadata hash mismatch for provider %s.", + cp->provider->name); + return (error); + } + return (0); +} + +/* + * Two functions below are responsible for updating metadata. + * Only metadata on the data provider is updated (we need to update + * information about active journal in there). + */ +static void +g_journal_metadata_done(struct bio *bp) +{ + + /* + * There is not much we can do on error except informing about it. + */ + if (bp->bio_error != 0) { + GJ_LOGREQ(0, bp, "Cannot update metadata (error=%d).", + bp->bio_error); + } else { + GJ_LOGREQ(2, bp, "Metadata updated."); + } + gj_free(bp->bio_data, bp->bio_length); + g_destroy_bio(bp); +} + +static void +g_journal_metadata_update(struct g_journal_softc *sc) +{ + struct g_journal_metadata md; + struct g_consumer *cp; + struct bio *bp; + u_char *sector; + + cp = sc->sc_dconsumer; + sector = gj_malloc(cp->provider->sectorsize, M_WAITOK); + strlcpy(md.md_magic, G_JOURNAL_MAGIC, sizeof(md.md_magic)); + md.md_version = G_JOURNAL_VERSION; + md.md_id = sc->sc_id; + md.md_type = sc->sc_orig_type; + md.md_jstart = sc->sc_jstart; + md.md_jend = sc->sc_jend; + md.md_joffset = sc->sc_inactive.jj_offset; + md.md_jid = sc->sc_journal_previous_id; + md.md_flags = 0; + if (sc->sc_flags & GJF_DEVICE_CLEAN) + md.md_flags |= GJ_FLAG_CLEAN; + + if (sc->sc_flags & GJF_DEVICE_HARDCODED) + strlcpy(md.md_provider, sc->sc_name, sizeof(md.md_provider)); + else + bzero(md.md_provider, sizeof(md.md_provider)); + md.md_provsize = cp->provider->mediasize; + journal_metadata_encode(&md, sector); + + bp = g_alloc_bio(); + bp->bio_offset = cp->provider->mediasize - cp->provider->sectorsize; + bp->bio_length = cp->provider->sectorsize; + bp->bio_data = sector; + bp->bio_cmd = BIO_WRITE; + if (!(sc->sc_flags & GJF_DEVICE_DESTROY)) { + bp->bio_done = g_journal_metadata_done; + g_io_request(bp, cp); + } else { + bp->bio_done = NULL; + g_io_request(bp, cp); + biowait(bp, "gjmdu"); + g_journal_metadata_done(bp); + } + g_journal_flush_cache(sc); +} + +/* + * This is where the I/O request comes from the GEOM. + */ +static void +g_journal_start(struct bio *bp) +{ + struct g_journal_softc *sc; + + sc = bp->bio_to->geom->softc; + GJ_LOGREQ(3, bp, "Request received."); + + switch (bp->bio_cmd) { + case BIO_READ: + case BIO_WRITE: + /* We support only those two requests. */ + break; + case BIO_DELETE: + case BIO_GETATTR: + default: + g_io_deliver(bp, EOPNOTSUPP); + return; + } + mtx_lock(&sc->sc_mtx); + bioq_insert_tail(&sc->sc_regular_queue, bp); + wakeup(sc); + mtx_unlock(&sc->sc_mtx); +} + +static void +g_journal_std_done(struct bio *bp) +{ + struct g_journal_softc *sc; + + sc = bp->bio_from->geom->softc; + mtx_lock(&sc->sc_mtx); + bioq_insert_tail(&sc->sc_back_queue, bp); + wakeup(sc); + mtx_unlock(&sc->sc_mtx); +} + +static struct bio * +g_journal_new_bio(off_t start, off_t end, off_t joffset, u_char *data, + int flags) +{ + struct bio *bp; + + bp = g_alloc_bio(); + bp->bio_offset = start; + bp->bio_joffset = joffset; + bp->bio_length = end - start; + bp->bio_cmd = BIO_WRITE; + bp->bio_done = g_journal_std_done; + if (data == NULL) + bp->bio_data = NULL; + else { + bp->bio_data = gj_malloc(bp->bio_length, flags); + if (bp->bio_data != NULL) + bcopy(data, bp->bio_data, bp->bio_length); + } + return (bp); +} + +#define g_journal_insert_bio(head, bp, flags) \ + g_journal_insert((head), (bp)->bio_offset, \ + (bp)->bio_offset + (bp)->bio_length, (bp)->bio_joffset, \ + (bp)->bio_data, flags) +/* + * The function below does a lot more than just inserting bio to the queue. + * It keeps the queue sorted by offset and ensures that there are no doubled + * data (it combines bios where ranges overlap). + * + * The function returns the number of bios inserted (as bio can be splitted). + */ +static int +g_journal_insert(struct bio **head, off_t nstart, off_t nend, off_t joffset, + u_char *data, int flags) +{ + struct bio *nbp, *cbp, *pbp; + off_t cstart, cend; + u_char *tmpdata; + int n; + + GJ_DEBUG(3, "INSERT(%p): (%jd, %jd, %jd)", *head, nstart, nend, + joffset); + n = 0; + pbp = NULL; + GJQ_FOREACH(*head, cbp) { + cstart = cbp->bio_offset; + cend = cbp->bio_offset + cbp->bio_length; + + if (nstart >= cend) { + /* + * +-------------+ + * | | + * | current | +-------------+ + * | bio | | | + * | | | new | + * +-------------+ | bio | + * | | + * +-------------+ + */ + GJ_DEBUG(3, "INSERT(%p): 1", *head); + } else if (nend <= cstart) { + /* + * +-------------+ + * | | + * +-------------+ | current | + * | | | bio | + * | new | | | + * | bio | +-------------+ + * | | + * +-------------+ + */ + nbp = g_journal_new_bio(nstart, nend, joffset, data, + flags); + if (pbp == NULL) + *head = nbp; + else + pbp->bio_next = nbp; + nbp->bio_next = cbp; + n++; + GJ_DEBUG(3, "INSERT(%p): 2 (nbp=%p pbp=%p)", *head, nbp, + pbp); + goto end; + } else if (nstart <= cstart && nend >= cend) { + /* + * +-------------+ +-------------+ + * | current bio | | current bio | + * +---+-------------+---+ +-------------+---+ + * | | | | | | | + * | | | | | | | + * | +-------------+ | +-------------+ | + * | new bio | | new bio | + * +---------------------+ +-----------------+ + * + * +-------------+ +-------------+ + * | current bio | | current bio | + * +---+-------------+ +-------------+ + * | | | | | + * | | | | | + * | +-------------+ +-------------+ + * | new bio | | new bio | + * +-----------------+ +-------------+ + */ + g_journal_stats_bytes_skipped += cbp->bio_length; + cbp->bio_offset = nstart; + cbp->bio_joffset = joffset; + cbp->bio_length = cend - nstart; + if (cbp->bio_data != NULL) { + gj_free(cbp->bio_data, cend - cstart); + cbp->bio_data = NULL; + } + if (data != NULL) { + cbp->bio_data = gj_malloc(cbp->bio_length, + flags); + if (cbp->bio_data != NULL) { + bcopy(data, cbp->bio_data, + cbp->bio_length); + } + data += cend - nstart; + } + joffset += cend - nstart; + nstart = cend; + GJ_DEBUG(3, "INSERT(%p): 3 (cbp=%p)", *head, cbp); + } else if (nstart > cstart && nend >= cend) { + /* + * +-----------------+ +-------------+ + * | current bio | | current bio | + * | +-------------+ | +---------+---+ + * | | | | | | | + * | | | | | | | + * +---+-------------+ +---+---------+ | + * | new bio | | new bio | + * +-------------+ +-------------+ + */ + g_journal_stats_bytes_skipped += cend - nstart; + nbp = g_journal_new_bio(nstart, cend, joffset, data, + flags); + nbp->bio_next = cbp->bio_next; + cbp->bio_next = nbp; + cbp->bio_length = nstart - cstart; + if (cbp->bio_data != NULL) { + cbp->bio_data = gj_realloc(cbp->bio_data, + cbp->bio_length, cend - cstart); + } + if (data != NULL) + data += cend - nstart; + joffset += cend - nstart; + nstart = cend; + n++; + GJ_DEBUG(3, "INSERT(%p): 4 (cbp=%p)", *head, cbp); + } else if (nstart > cstart && nend < cend) { + /* + * +---------------------+ + * | current bio | + * | +-------------+ | + * | | | | + * | | | | + * +---+-------------+---+ + * | new bio | + * +-------------+ + */ + g_journal_stats_bytes_skipped += nend - nstart; + nbp = g_journal_new_bio(nstart, nend, joffset, data, + flags); + nbp->bio_next = cbp->bio_next; + cbp->bio_next = nbp; + if (cbp->bio_data == NULL) + tmpdata = NULL; + else + tmpdata = cbp->bio_data + nend - cstart; + nbp = g_journal_new_bio(nend, cend, + cbp->bio_joffset + nend - cstart, tmpdata, flags); + nbp->bio_next = ((struct bio *)cbp->bio_next)->bio_next; + ((struct bio *)cbp->bio_next)->bio_next = nbp; + cbp->bio_length = nstart - cstart; + if (cbp->bio_data != NULL) { + cbp->bio_data = gj_realloc(cbp->bio_data, + cbp->bio_length, cend - cstart); + } + n += 2; + GJ_DEBUG(3, "INSERT(%p): 5 (cbp=%p)", *head, cbp); + goto end; + } else if (nstart <= cstart && nend < cend) { + /* + * +-----------------+ +-------------+ + * | current bio | | current bio | + * +-------------+ | +---+---------+ | + * | | | | | | | + * | | | | | | | + * +-------------+---+ | +---------+---+ + * | new bio | | new bio | + * +-------------+ +-------------+ + */ + g_journal_stats_bytes_skipped += nend - nstart; + nbp = g_journal_new_bio(nstart, nend, joffset, data, + flags); + if (pbp == NULL) + *head = nbp; + else + pbp->bio_next = nbp; + nbp->bio_next = cbp; + cbp->bio_offset = nend; + cbp->bio_length = cend - nend; + cbp->bio_joffset += nend - cstart; + tmpdata = cbp->bio_data; + if (tmpdata != NULL) { + cbp->bio_data = gj_malloc(cbp->bio_length, + flags); + if (cbp->bio_data != NULL) { + bcopy(tmpdata + nend - cstart, + cbp->bio_data, cbp->bio_length); + } + gj_free(tmpdata, cend - cstart); + } + n++; + GJ_DEBUG(3, "INSERT(%p): 6 (cbp=%p)", *head, cbp); + goto end; + } + if (nstart == nend) + goto end; + pbp = cbp; + } + nbp = g_journal_new_bio(nstart, nend, joffset, data, flags); + if (pbp == NULL) + *head = nbp; + else + pbp->bio_next = nbp; + nbp->bio_next = NULL; + n++; + GJ_DEBUG(3, "INSERT(%p): 8 (nbp=%p pbp=%p)", *head, nbp, pbp); +end: + if (g_journal_debug >= 3) { + GJQ_FOREACH(*head, cbp) { + GJ_DEBUG(3, "ELEMENT: %p (%jd, %jd, %jd, %p)", cbp, + (intmax_t)cbp->bio_offset, + (intmax_t)cbp->bio_length, + (intmax_t)cbp->bio_joffset, cbp->bio_data); + } + GJ_DEBUG(3, "INSERT(%p): DONE %d", *head, n); + } + return (n); +} + +/* + * The function combines neighbour bios trying to squeeze as much data as + * possible into one bio. + * + * The function returns the number of bios combined (negative value). + */ +static int +g_journal_optimize(struct bio *head) +{ + struct bio *cbp, *pbp; + int n; + + n = 0; + pbp = NULL; + GJQ_FOREACH(head, cbp) { + /* Skip bios which has to be read first. */ + if (cbp->bio_data == NULL) { + pbp = NULL; + continue; + } + /* There is no previous bio yet. */ + if (pbp == NULL) { + pbp = cbp; + continue; + } + /* Is this a neighbour bio? */ + if (pbp->bio_offset + pbp->bio_length != cbp->bio_offset) { + /* Be sure that bios queue is sorted. */ + KASSERT(pbp->bio_offset + pbp->bio_length < cbp->bio_offset, + ("poffset=%jd plength=%jd coffset=%jd", + (intmax_t)pbp->bio_offset, + (intmax_t)pbp->bio_length, + (intmax_t)cbp->bio_offset)); + pbp = cbp; + continue; + } + /* Be sure we don't end up with too big bio. */ + if (pbp->bio_length + cbp->bio_length > MAXPHYS) { + pbp = cbp; + continue; + } + /* Ok, we can join bios. */ + GJ_LOGREQ(4, pbp, "Join: "); + GJ_LOGREQ(4, cbp, "and: "); + pbp->bio_data = gj_realloc(pbp->bio_data, + pbp->bio_length + cbp->bio_length, pbp->bio_length); + bcopy(cbp->bio_data, pbp->bio_data + pbp->bio_length, + cbp->bio_length); + gj_free(cbp->bio_data, cbp->bio_length); + pbp->bio_length += cbp->bio_length; + pbp->bio_next = cbp->bio_next; + g_destroy_bio(cbp); + cbp = pbp; + g_journal_stats_combined_ios++; + n--; + GJ_LOGREQ(4, pbp, "Got: "); + } + return (n); +} + +/* + * TODO: Update comment. + * These are functions responsible for copying one portion of data from journal + * to the destination provider. + * The order goes like this: + * 1. Read the header, which contains informations about data blocks + * following it. + * 2. Read the data blocks from the journal. + * 3. Write the data blocks on the data provider. + * + * g_journal_copy_start() + * g_journal_copy_done() - got finished write request, logs potential errors. + */ + +/* + * When there is no data in cache, this function is used to read it. + */ +static void +g_journal_read_first(struct g_journal_softc *sc, struct bio *bp) +{ + struct bio *cbp; + + /* + * We were short in memory, so data was freed. + * In that case we need to read it back from journal. + */ + cbp = g_alloc_bio(); + cbp->bio_cflags = bp->bio_cflags; + cbp->bio_parent = bp; + cbp->bio_offset = bp->bio_joffset; + cbp->bio_length = bp->bio_length; + cbp->bio_data = gj_malloc(bp->bio_length, M_WAITOK); + cbp->bio_cmd = BIO_READ; + cbp->bio_done = g_journal_std_done; + GJ_LOGREQ(4, cbp, "READ FIRST"); + g_io_request(cbp, sc->sc_jconsumer); + g_journal_cache_misses++; +} + +static void +g_journal_copy_send(struct g_journal_softc *sc) +{ + struct bio *bioq, *bp, *lbp; + + bioq = lbp = NULL; + mtx_lock(&sc->sc_mtx); + for (; sc->sc_copy_in_progress < g_journal_parallel_copies;) { + bp = GJQ_FIRST(sc->sc_inactive.jj_queue); + if (bp == NULL) + break; + GJQ_REMOVE(sc->sc_inactive.jj_queue, bp); + sc->sc_copy_in_progress++; + GJQ_INSERT_AFTER(bioq, bp, lbp); + lbp = bp; + } + mtx_unlock(&sc->sc_mtx); + if (g_journal_do_optimize) + sc->sc_copy_in_progress += g_journal_optimize(bioq); + while ((bp = GJQ_FIRST(bioq)) != NULL) { + GJQ_REMOVE(bioq, bp); + GJQ_INSERT_HEAD(sc->sc_copy_queue, bp); + bp->bio_cflags = GJ_BIO_COPY; + if (bp->bio_data == NULL) + g_journal_read_first(sc, bp); + else { + bp->bio_joffset = 0; + GJ_LOGREQ(4, bp, "SEND"); + g_io_request(bp, sc->sc_dconsumer); + } + } +} + +static void +g_journal_copy_start(struct g_journal_softc *sc) +{ + + /* + * Remember in metadata that we're starting to copy journaled data + * to the data provider. + * In case of power failure, we will copy these data once again on boot. + */ + if (!sc->sc_journal_copying) { + sc->sc_journal_copying = 1; + GJ_DEBUG(1, "Starting copy of journal."); + g_journal_metadata_update(sc); + } + g_journal_copy_send(sc); +} + +/* + * Data block has been read from the journal provider. + */ +static int +g_journal_copy_read_done(struct bio *bp) +{ + struct g_journal_softc *sc; + struct g_consumer *cp; + struct bio *pbp; + + KASSERT(bp->bio_cflags == GJ_BIO_COPY, + ("Invalid bio (%d != %d).", bp->bio_cflags, GJ_BIO_COPY)); + + sc = bp->bio_from->geom->softc; + pbp = bp->bio_parent; + + if (bp->bio_error != 0) { + GJ_DEBUG(0, "Error while reading data from %s (error=%d).", + bp->bio_to->name, bp->bio_error); + /* + * We will not be able to deliver WRITE request as well. + */ + gj_free(bp->bio_data, bp->bio_length); + g_destroy_bio(pbp); + g_destroy_bio(bp); + sc->sc_copy_in_progress--; + return (1); + } + pbp->bio_data = bp->bio_data; + cp = sc->sc_dconsumer; + g_io_request(pbp, cp); + GJ_LOGREQ(4, bp, "READ DONE"); + g_destroy_bio(bp); + return (0); +} + +/* + * Data block has been written to the data provider. + */ +static void +g_journal_copy_write_done(struct bio *bp) +{ + struct g_journal_softc *sc; + + KASSERT(bp->bio_cflags == GJ_BIO_COPY, + ("Invalid bio (%d != %d).", bp->bio_cflags, GJ_BIO_COPY)); + + sc = bp->bio_from->geom->softc; + sc->sc_copy_in_progress--; + + if (bp->bio_error != 0) { + GJ_LOGREQ(0, bp, "[copy] Error while writting data (error=%d)", + bp->bio_error); + } + GJQ_REMOVE(sc->sc_copy_queue, bp); + gj_free(bp->bio_data, bp->bio_length); + GJ_LOGREQ(4, bp, "DONE"); + g_destroy_bio(bp); + + if (sc->sc_copy_in_progress == 0) { + /* + * This was the last write request for this journal. + */ + GJ_DEBUG(1, "Data has been copied."); + sc->sc_journal_copying = 0; + } +} + +static void g_journal_flush_done(struct bio *bp); + +/* + * Flush one record onto active journal provider. + */ +static void +g_journal_flush(struct g_journal_softc *sc) +{ + struct g_journal_record_header hdr; + struct g_journal_entry *ent; + struct g_provider *pp; + struct bio **bioq; + struct bio *bp, *fbp, *pbp; + off_t joffset, size; + u_char *data, hash[16]; + MD5_CTX ctx; + u_int i; + + if (sc->sc_current_count == 0) + return; + + size = 0; + pp = sc->sc_jprovider; + GJ_VALIDATE_OFFSET(sc->sc_journal_offset, sc); + joffset = sc->sc_journal_offset; + + GJ_DEBUG(2, "Storing %d journal entries on %s at %jd.", + sc->sc_current_count, pp->name, (intmax_t)joffset); + + /* + * Store 'journal id', so we know to which journal this record belongs. + */ + hdr.jrh_journal_id = sc->sc_journal_id; + /* Could be less than g_journal_record_entries if called due timeout. */ + hdr.jrh_nentries = MIN(sc->sc_current_count, g_journal_record_entries); + strlcpy(hdr.jrh_magic, GJ_RECORD_HEADER_MAGIC, sizeof(hdr.jrh_magic)); + + bioq = &sc->sc_active.jj_queue; + pbp = sc->sc_flush_queue; + + fbp = g_alloc_bio(); + fbp->bio_parent = NULL; + fbp->bio_cflags = GJ_BIO_JOURNAL; + fbp->bio_offset = -1; + fbp->bio_joffset = joffset; + fbp->bio_length = pp->sectorsize; + fbp->bio_cmd = BIO_WRITE; + fbp->bio_done = g_journal_std_done; + GJQ_INSERT_AFTER(sc->sc_flush_queue, fbp, pbp); + pbp = fbp; + fbp->bio_to = pp; + GJ_LOGREQ(4, fbp, "FLUSH_OUT"); + joffset += pp->sectorsize; + sc->sc_flush_count++; + if (sc->sc_flags & GJF_DEVICE_CHECKSUM) + MD5Init(&ctx); + + for (i = 0; i < hdr.jrh_nentries; i++) { + bp = sc->sc_current_queue; + KASSERT(bp != NULL, ("NULL bp")); + bp->bio_to = pp; + GJ_LOGREQ(4, bp, "FLUSHED"); + sc->sc_current_queue = bp->bio_next; + bp->bio_next = NULL; + sc->sc_current_count--; + + /* Add to the header. */ + ent = &hdr.jrh_entries[i]; + ent->je_offset = bp->bio_offset; + ent->je_joffset = joffset; + ent->je_length = bp->bio_length; + size += ent->je_length; + + data = bp->bio_data; + if (sc->sc_flags & GJF_DEVICE_CHECKSUM) + MD5Update(&ctx, data, ent->je_length); + bzero(bp, sizeof(*bp)); + bp->bio_cflags = GJ_BIO_JOURNAL; + bp->bio_offset = ent->je_offset; + bp->bio_joffset = ent->je_joffset; + bp->bio_length = ent->je_length; + bp->bio_data = data; + bp->bio_cmd = BIO_WRITE; + bp->bio_done = g_journal_std_done; + GJQ_INSERT_AFTER(sc->sc_flush_queue, bp, pbp); + pbp = bp; + bp->bio_to = pp; + GJ_LOGREQ(4, bp, "FLUSH_OUT"); + joffset += bp->bio_length; + sc->sc_flush_count++; + + /* + * Add request to the active sc_journal_queue queue. + * This is our cache. After journal switch we don't have to + * read the data from the inactive journal, because we keep + * it in memory. + */ + g_journal_insert(bioq, ent->je_offset, + ent->je_offset + ent->je_length, ent->je_joffset, data, + M_NOWAIT); + } + + /* + * After all requests, store valid header. + */ + data = gj_malloc(pp->sectorsize, M_WAITOK); + if (sc->sc_flags & GJF_DEVICE_CHECKSUM) { + MD5Final(hash, &ctx); + bcopy(hash, hdr.jrh_sum, sizeof(hdr.jrh_sum)); + } + g_journal_record_header_encode(&hdr, data); + fbp->bio_data = data; + + sc->sc_journal_offset = joffset; + if ((sc->sc_active.jj_offset < sc->sc_inactive.jj_offset && + sc->sc_journal_offset >= sc->sc_inactive.jj_offset) || + (sc->sc_active.jj_offset > sc->sc_inactive.jj_offset && + sc->sc_journal_offset >= sc->sc_inactive.jj_offset && + sc->sc_journal_offset < sc->sc_active.jj_offset)) { + panic("Journal overflow (joffset=%jd active=%jd inactive=%jd)", + (intmax_t)sc->sc_journal_offset, + (intmax_t)sc->sc_active.jj_offset, + (intmax_t)sc->sc_inactive.jj_offset); + } +} + +/* + * Flush request finished. + */ +static void +g_journal_flush_done(struct bio *bp) +{ + struct g_journal_softc *sc; + struct g_consumer *cp; + + KASSERT((bp->bio_cflags & GJ_BIO_MASK) == GJ_BIO_JOURNAL, + ("Invalid bio (%d != %d).", bp->bio_cflags, GJ_BIO_JOURNAL)); + + cp = bp->bio_from; + sc = cp->geom->softc; + sc->sc_flush_in_progress--; + + if (bp->bio_error != 0) { + GJ_LOGREQ(0, bp, "[flush] Error while writting data (error=%d)", + bp->bio_error); + } + gj_free(bp->bio_data, bp->bio_length); + GJ_LOGREQ(4, bp, "DONE"); + g_destroy_bio(bp); +} + +static void g_journal_add_request(struct g_journal_softc *sc, struct bio *bp); + +static void +g_journal_flush_send(struct g_journal_softc *sc) +{ + struct g_consumer *cp; + struct bio *bioq, *bp1, *bp2, *lbp; + + cp = sc->sc_jconsumer; + bioq = lbp = NULL; + for (; sc->sc_flush_in_progress < g_journal_parallel_flushes; ) { + /* Send one flush requests to the active journal. */ + bp1 = GJQ_FIRST(sc->sc_flush_queue); + if (bp1 != NULL) { + GJQ_REMOVE(sc->sc_flush_queue, bp1); + sc->sc_flush_count--; + bp1->bio_offset = bp1->bio_joffset; + bp1->bio_joffset = 0; + sc->sc_flush_in_progress++; + GJQ_INSERT_AFTER(bioq, bp1, lbp); + lbp = bp1; + } + /* Take one delayed requests as we have free place now. */ + bp2 = bioq_takefirst(&sc->sc_delayed_queue); + if (bp2 != NULL) { + sc->sc_delayed_count--; + GJ_LOGREQ(4, bp2, "Delayed->Current"); + g_journal_add_request(sc, bp2); + } + /* + * If there are no requests to flush and no delayed requests + * there is nothing to do, leave. + */ + /* XXX: Is this correct? Do we want to take all delayed requests? */ + if (bp1 == NULL && bp2 == NULL) + break; + } + if (g_journal_do_optimize) + sc->sc_flush_in_progress += g_journal_optimize(bioq); + while ((bp1 = GJQ_FIRST(bioq)) != NULL) { + GJQ_REMOVE(bioq, bp1); + GJ_LOGREQ(3, bp1, "Flush request send"); + g_io_request(bp1, cp); + } +} + +/* + * Add I/O request to the current queue. If we have enough requests for one + * journal record we flush them onto active journal provider. + */ +static void +g_journal_add_request(struct g_journal_softc *sc, struct bio *bp) +{ + int n; + + /* + * The flush queue is full, we need to delay the request. + */ + if (sc->sc_flush_count >= g_journal_accept_immediately) { + GJ_LOGREQ(4, bp, "DELAYED"); + bioq_insert_tail(&sc->sc_delayed_queue, bp); + sc->sc_delayed_count++; + return; + } + + GJ_LOGREQ(4, bp, "CURRENT %d", sc->sc_current_count); + n = g_journal_insert_bio(&sc->sc_current_queue, bp, M_WAITOK); + sc->sc_current_count += n; + n = g_journal_optimize(sc->sc_current_queue); + sc->sc_current_count += n; + /* + * For requests which are added to the current queue we deliver + * response immediately. + */ + bp->bio_completed = bp->bio_length; + g_io_deliver(bp, 0); + if (sc->sc_current_count >= g_journal_record_entries) { + /* + * Let's flush one record onto active journal provider. + */ + g_journal_flush(sc); + } +} + +static void g_journal_read_done(struct bio *bp); + +/* + * Try to find requested data in cache. + */ +static struct bio * +g_journal_read_find(struct bio *head, int sorted, struct bio *pbp, off_t ostart, + off_t oend) +{ + off_t cstart, cend; + struct bio *bp; + + GJQ_FOREACH(head, bp) { + if (bp->bio_offset == -1) + continue; + cstart = MAX(ostart, bp->bio_offset); + cend = MIN(oend, bp->bio_offset + bp->bio_length); + if (cend <= ostart) + continue; + else if (cstart >= oend) { + if (!sorted) + continue; + else { + bp = NULL; + break; + } + } + if (bp->bio_data == NULL) + break; + GJ_DEBUG(3, "READ(%p): (%jd, %jd) (bp=%p)", head, cstart, cend, + bp); + bcopy(bp->bio_data + cstart - bp->bio_offset, + pbp->bio_data + cstart - pbp->bio_offset, cend - cstart); + pbp->bio_completed += cend - cstart; + if (pbp->bio_completed == pbp->bio_length) { + /* + * Cool, the whole request was in cache, deliver happy + * message. + */ + g_io_deliver(pbp, 0); + return (pbp); + } + break; + } + return (bp); +} + +/* + * Try to find requested data in cache. + */ +static struct bio * +g_journal_read_queue_find(struct bio_queue *head, struct bio *pbp, off_t ostart, + off_t oend) +{ + off_t cstart, cend; + struct bio *bp; + + TAILQ_FOREACH(bp, head, bio_queue) { + cstart = MAX(ostart, bp->bio_offset); + cend = MIN(oend, bp->bio_offset + bp->bio_length); + if (cend <= ostart) + continue; + else if (cstart >= oend) + continue; + KASSERT(bp->bio_data != NULL, + ("%s: bio_data == NULL", __func__)); + GJ_DEBUG(3, "READ(%p): (%jd, %jd) (bp=%p)", head, cstart, cend, + bp); + bcopy(bp->bio_data + cstart - bp->bio_offset, + pbp->bio_data + cstart - pbp->bio_offset, cend - cstart); + pbp->bio_completed += cend - cstart; + if (pbp->bio_completed == pbp->bio_length) { + /* + * Cool, the whole request was in cache, deliver happy + * message. + */ + g_io_deliver(pbp, 0); + return (pbp); + } + break; + } + return (bp); +} + +/* + * This function is used for colecting data on read. + * The complexity is because parts of the data can be stored in four different + * places: + * - in delayed requests + * - in memory - the data not yet send to the active journal provider + * - in requests which are going to be sent to the active journal + * - in the active journal + * - in the inactive journal + * - in the data provider + */ +static void +g_journal_read(struct g_journal_softc *sc, struct bio *pbp, off_t ostart, + off_t oend) +{ + struct bio *bp, *nbp, *head; + off_t cstart, cend; + u_int i, sorted = 0; + + GJ_DEBUG(3, "READ: (%jd, %jd)", ostart, oend); + + cstart = cend = -1; + bp = NULL; + head = NULL; + for (i = 0; i <= 5; i++) { + switch (i) { + case 0: /* Delayed requests. */ + head = NULL; + sorted = 0; + break; + case 1: /* Not-yet-send data. */ + head = sc->sc_current_queue; + sorted = 1; + break; + case 2: /* In-flight to the active journal. */ + head = sc->sc_flush_queue; + sorted = 0; + break; + case 3: /* Active journal. */ + head = sc->sc_active.jj_queue; + sorted = 1; + break; + case 4: /* Inactive journal. */ + /* + * XXX: Here could be a race with g_journal_lowmem(). + */ + head = sc->sc_inactive.jj_queue; + sorted = 1; + break; + case 5: /* In-flight to the data provider. */ + head = sc->sc_copy_queue; + sorted = 0; + break; + default: + panic("gjournal %s: i=%d", __func__, i); + } + if (i == 0) + bp = g_journal_read_queue_find(&sc->sc_delayed_queue.queue, pbp, ostart, oend); + else + bp = g_journal_read_find(head, sorted, pbp, ostart, oend); + if (bp == pbp) { /* Got the whole request. */ + GJ_DEBUG(2, "Got the whole request from %u.", i); + return; + } else if (bp != NULL) { + cstart = MAX(ostart, bp->bio_offset); + cend = MIN(oend, bp->bio_offset + bp->bio_length); + GJ_DEBUG(2, "Got part of the request from %u (%jd-%jd).", + i, (intmax_t)cstart, (intmax_t)cend); + break; + } + } + if (bp != NULL) { + if (bp->bio_data == NULL) { + nbp = g_clone_bio(pbp); + nbp->bio_cflags = GJ_BIO_READ; + nbp->bio_data = + pbp->bio_data + cstart - pbp->bio_offset; + nbp->bio_offset = + bp->bio_joffset + cstart - bp->bio_offset; + nbp->bio_length = cend - cstart; + nbp->bio_done = g_journal_read_done; + g_io_request(nbp, sc->sc_jconsumer); + } + /* + * If we don't have the whole request yet, call g_journal_read() + * recursively. + */ + if (ostart < cstart) + g_journal_read(sc, pbp, ostart, cstart); + if (oend > cend) + g_journal_read(sc, pbp, cend, oend); + } else { + /* + * No data in memory, no data in journal. + * Its time for asking data provider. + */ + GJ_DEBUG(3, "READ(data): (%jd, %jd)", ostart, oend); + nbp = g_clone_bio(pbp); + nbp->bio_cflags = GJ_BIO_READ; + nbp->bio_data = pbp->bio_data + ostart - pbp->bio_offset; + nbp->bio_offset = ostart; + nbp->bio_length = oend - ostart; + nbp->bio_done = g_journal_read_done; + g_io_request(nbp, sc->sc_dconsumer); + /* We have the whole request, return here. */ + return; + } +} + +/* + * Function responsible for handling finished READ requests. + * Actually, g_std_done() could be used here, the only difference is that we + * log error. + */ +static void +g_journal_read_done(struct bio *bp) +{ + struct bio *pbp; + + KASSERT(bp->bio_cflags == GJ_BIO_READ, + ("Invalid bio (%d != %d).", bp->bio_cflags, GJ_BIO_READ)); + + pbp = bp->bio_parent; + pbp->bio_inbed++; + pbp->bio_completed += bp->bio_length; + + if (bp->bio_error != 0) { + if (pbp->bio_error == 0) + pbp->bio_error = bp->bio_error; + GJ_DEBUG(0, "Error while reading data from %s (error=%d).", + bp->bio_to->name, bp->bio_error); + } + g_destroy_bio(bp); + if (pbp->bio_children == pbp->bio_inbed && + pbp->bio_completed == pbp->bio_length) { + /* We're done. */ + g_io_deliver(pbp, 0); + } +} + +/* + * Deactive current journal and active next one. + */ +static void +g_journal_switch(struct g_journal_softc *sc) +{ + struct g_provider *pp; + + if (JEMPTY(sc)) { + GJ_DEBUG(3, "No need for %s switch.", sc->sc_name); + pp = LIST_FIRST(&sc->sc_geom->provider); + if (!(sc->sc_flags & GJF_DEVICE_CLEAN) && pp->acw == 0) { + sc->sc_flags |= GJF_DEVICE_CLEAN; + GJ_DEBUG(1, "Marking %s as clean.", sc->sc_name); + g_journal_metadata_update(sc); + } + } else { + GJ_DEBUG(3, "Switching journal %s.", sc->sc_geom->name); + + pp = sc->sc_jprovider; + + sc->sc_journal_previous_id = sc->sc_journal_id; + + sc->sc_journal_id = sc->sc_journal_next_id; + sc->sc_journal_next_id = arc4random(); + + GJ_VALIDATE_OFFSET(sc->sc_journal_offset, sc); + + g_journal_write_header(sc); + + sc->sc_inactive.jj_offset = sc->sc_active.jj_offset; + sc->sc_inactive.jj_queue = sc->sc_active.jj_queue; + + sc->sc_active.jj_offset = + sc->sc_journal_offset - pp->sectorsize; + sc->sc_active.jj_queue = NULL; + + /* + * Switch is done, start copying data from the (now) inactive + * journal to the data provider. + */ + g_journal_copy_start(sc); + } + mtx_lock(&sc->sc_mtx); + sc->sc_flags &= ~GJF_DEVICE_SWITCH; + mtx_unlock(&sc->sc_mtx); +} + +static void +g_journal_initialize(struct g_journal_softc *sc) +{ + + sc->sc_journal_id = arc4random(); + sc->sc_journal_next_id = arc4random(); + sc->sc_journal_previous_id = sc->sc_journal_id; + sc->sc_journal_offset = sc->sc_jstart; + sc->sc_inactive.jj_offset = sc->sc_jstart; + g_journal_write_header(sc); + sc->sc_active.jj_offset = sc->sc_jstart; +} + +static void +g_journal_mark_as_dirty(struct g_journal_softc *sc) +{ + const struct g_journal_desc *desc; + int i; + + GJ_DEBUG(1, "Marking file system %s as dirty.", sc->sc_name); + for (i = 0; (desc = g_journal_filesystems[i]) != NULL; i++) + desc->jd_dirty(sc->sc_dconsumer); +} + +/* + * Function read record header from the given journal. + * It is very simlar to g_read_data(9), but it doesn't allocate memory for bio + * and data on every call. + */ +static int +g_journal_sync_read(struct g_consumer *cp, struct bio *bp, off_t offset, + void *data) +{ + int error; + + bzero(bp, sizeof(*bp)); + bp->bio_cmd = BIO_READ; + bp->bio_done = NULL; + bp->bio_offset = offset; + bp->bio_length = cp->provider->sectorsize; + bp->bio_data = data; + g_io_request(bp, cp); + error = biowait(bp, "gjs_read"); + return (error); +} + +#if 0 +/* + * Function is called when we start the journal device and we detect that + * one of the journals was not fully copied. + * The purpose of this function is to read all records headers from journal + * and placed them in the inactive queue, so we can start journal + * synchronization process and the journal provider itself. + * Design decision was taken to not synchronize the whole journal here as it + * can take too much time. Reading headers only and delaying synchronization + * process until after journal provider is started should be the best choice. + */ +#endif + +static void +g_journal_sync(struct g_journal_softc *sc) +{ + struct g_journal_record_header rhdr; + struct g_journal_entry *ent; + struct g_journal_header jhdr; + struct g_consumer *cp; + struct bio *bp, *fbp, *tbp; + off_t joffset, offset; + u_char *buf, sum[16]; + uint64_t id; + MD5_CTX ctx; + int error, found, i; + + found = 0; + fbp = NULL; + cp = sc->sc_jconsumer; + bp = g_alloc_bio(); + buf = gj_malloc(cp->provider->sectorsize, M_WAITOK); + joffset = sc->sc_journal_offset; + offset = joffset; + sc->sc_inactive.jj_offset = joffset; + + GJ_DEBUG(2, "Looking for termination at %jd.", (intmax_t)joffset); + + /* + * Read and decode first journal header. + */ + error = g_journal_sync_read(cp, bp, offset, buf); + if (error != 0) { + GJ_DEBUG(0, "Error while reading journal header from %s.", + cp->provider->name); + goto end; + } + error = g_journal_header_decode(buf, &jhdr); + if (error != 0) { + GJ_DEBUG(0, "Cannot decode journal header from %s.", + cp->provider->name); + goto end; + } + id = sc->sc_journal_id; + if (jhdr.jh_journal_id != sc->sc_journal_id) { + GJ_DEBUG(1, "Journal ID mismatch at %jd (0x%08x != 0x%08x).", + (intmax_t)offset, (u_int)jhdr.jh_journal_id, (u_int)id); + goto end; + } + offset += cp->provider->sectorsize; + id = sc->sc_journal_next_id = jhdr.jh_journal_next_id; + + for (;;) { + /* + * If the biggest record won't fit, look for a record header or + * journal header from the begining. + */ + GJ_VALIDATE_OFFSET(offset, sc); + error = g_journal_sync_read(cp, bp, offset, buf); + if (error != 0) { + /* + * Not good. Having an error while reading header + * means, that we cannot read next headers and in + * consequence we cannot find termination. + */ + GJ_DEBUG(0, + "Error while reading record header from %s.", + cp->provider->name); + break; + } + + error = g_journal_record_header_decode(buf, &rhdr); + if (error != 0) { + GJ_DEBUG(2, "Not a record header at %jd (error=%d).", + (intmax_t)offset, error); + /* + * This is not a record header. + * If we are lucky, this is next journal header. + */ + error = g_journal_header_decode(buf, &jhdr); + if (error != 0) { + GJ_DEBUG(1, "Not a journal header at %jd (error=%d).", + (intmax_t)offset, error); + /* + * Nope, this is not journal header, which + * bascially means that journal is not + * terminated properly. + */ + error = ENOENT; + break; + } + /* + * Ok. This is header of _some_ journal. Now we need to + * verify if this is header of the _next_ journal. + */ + if (jhdr.jh_journal_id != id) { + GJ_DEBUG(1, "Journal ID mismatch at %jd " + "(0x%08x != 0x%08x).", (intmax_t)offset, + (u_int)jhdr.jh_journal_id, (u_int)id); + error = ENOENT; + break; + } + + /* Found termination. */ + found++; + GJ_DEBUG(1, "Found termination at %jd (id=0x%08x).", + (intmax_t)offset, (u_int)id); + sc->sc_active.jj_offset = offset; + sc->sc_journal_offset = + offset + cp->provider->sectorsize; + sc->sc_journal_id = id; + id = sc->sc_journal_next_id = jhdr.jh_journal_next_id; + + while ((tbp = fbp) != NULL) { + fbp = tbp->bio_next; + GJ_LOGREQ(3, tbp, "Adding request."); + g_journal_insert_bio(&sc->sc_inactive.jj_queue, + tbp, M_WAITOK); + } + + /* Skip journal's header. */ + offset += cp->provider->sectorsize; + continue; + } + + /* Skip record's header. */ + offset += cp->provider->sectorsize; + + /* + * Add information about every record entry to the inactive + * queue. + */ + if (sc->sc_flags & GJF_DEVICE_CHECKSUM) + MD5Init(&ctx); + for (i = 0; i < rhdr.jrh_nentries; i++) { + ent = &rhdr.jrh_entries[i]; + GJ_DEBUG(3, "Insert entry: %jd %jd.", + (intmax_t)ent->je_offset, (intmax_t)ent->je_length); + g_journal_insert(&fbp, ent->je_offset, + ent->je_offset + ent->je_length, ent->je_joffset, + NULL, M_WAITOK); + if (sc->sc_flags & GJF_DEVICE_CHECKSUM) { + u_char *buf2; + + /* + * TODO: Should use faster function (like + * g_journal_sync_read()). + */ + buf2 = g_read_data(cp, offset, ent->je_length, + NULL); + if (buf2 == NULL) + GJ_DEBUG(0, "Cannot read data at %jd.", + (intmax_t)offset); + else { + MD5Update(&ctx, buf2, ent->je_length); + g_free(buf2); + } + } + /* Skip entry's data. */ + offset += ent->je_length; + } + if (sc->sc_flags & GJF_DEVICE_CHECKSUM) { + MD5Final(sum, &ctx); + if (bcmp(sum, rhdr.jrh_sum, sizeof(rhdr.jrh_sum)) != 0) { + GJ_DEBUG(0, "MD5 hash mismatch at %jd!", + (intmax_t)offset); + } + } + } +end: + gj_free(bp->bio_data, cp->provider->sectorsize); + g_destroy_bio(bp); + + /* Remove bios from unterminated journal. */ + while ((tbp = fbp) != NULL) { + fbp = tbp->bio_next; + g_destroy_bio(tbp); + } + + if (found < 1) { + GJ_DEBUG(0, "Journal on %s is broken/corrupted. Initializing.", + sc->sc_name); + while ((tbp = sc->sc_inactive.jj_queue) != NULL) { + sc->sc_inactive.jj_queue = tbp->bio_next; + g_destroy_bio(tbp); + } + g_journal_initialize(sc); + g_journal_mark_as_dirty(sc); + } else { + GJ_DEBUG(0, "Journal %s consistent.", sc->sc_name); + g_journal_copy_start(sc); + } +} + +/* + * Wait for requests. + * If we have requests in the current queue, flush them after 3 seconds from the + * last flush. In this way we don't wait forever (or for journal switch) with + * storing not full records on journal. + */ +static void +g_journal_wait(struct g_journal_softc *sc, time_t last_write) +{ + int error, timeout; + + GJ_DEBUG(3, "%s: enter", __func__); + if (sc->sc_current_count == 0) { + if (g_journal_debug < 2) + msleep(sc, &sc->sc_mtx, PRIBIO | PDROP, "gj:work", 0); + else { + /* + * If we have debug turned on, show number of elements + * in various queues. + */ + for (;;) { + error = msleep(sc, &sc->sc_mtx, PRIBIO, + "gj:work", hz * 3); + if (error == 0) { + mtx_unlock(&sc->sc_mtx); + break; + } + GJ_DEBUG(3, "Report: current count=%d", + sc->sc_current_count); + GJ_DEBUG(3, "Report: flush count=%d", + sc->sc_flush_count); + GJ_DEBUG(3, "Report: flush in progress=%d", + sc->sc_flush_in_progress); + GJ_DEBUG(3, "Report: copy in progress=%d", + sc->sc_copy_in_progress); + GJ_DEBUG(3, "Report: delayed=%d", + sc->sc_delayed_count); + } + } + GJ_DEBUG(3, "%s: exit 1", __func__); + return; + } + + /* + * Flush even not full records every 3 seconds. + */ + timeout = (last_write + 3 - time_second) * hz; + if (timeout <= 0) { + mtx_unlock(&sc->sc_mtx); + g_journal_flush(sc); + g_journal_flush_send(sc); + GJ_DEBUG(3, "%s: exit 2", __func__); + return; + } + error = msleep(sc, &sc->sc_mtx, PRIBIO | PDROP, "gj:work", timeout); + if (error == EWOULDBLOCK) + g_journal_flush_send(sc); + GJ_DEBUG(3, "%s: exit 3", __func__); +} + +/* + * Worker thread. + */ +static void +g_journal_worker(void *arg) +{ + struct g_journal_softc *sc; + struct g_geom *gp; + struct g_provider *pp; + struct bio *bp; + time_t last_write; + int type; + + mtx_lock_spin(&sched_lock); + sched_prio(curthread, PRIBIO); + mtx_unlock_spin(&sched_lock); + + sc = arg; + + if (sc->sc_flags & GJF_DEVICE_CLEAN) { + GJ_DEBUG(0, "Journal %s clean.", sc->sc_name); + g_journal_initialize(sc); + } else { + g_journal_sync(sc); + } + /* + * Check if we can use BIO_FLUSH. + */ + sc->sc_bio_flush = 0; + if (g_io_flush(sc->sc_jconsumer) == 0) { + sc->sc_bio_flush |= GJ_FLUSH_JOURNAL; + GJ_DEBUG(0, "BIO_FLUSH supported by %s.", + sc->sc_jconsumer->provider->name); + } else { + GJ_DEBUG(0, "BIO_FLUSH not supported by %s.", + sc->sc_jconsumer->provider->name); + } + if (sc->sc_jconsumer != sc->sc_dconsumer) { + if (g_io_flush(sc->sc_dconsumer) == 0) { + sc->sc_bio_flush |= GJ_FLUSH_DATA; + GJ_DEBUG(0, "BIO_FLUSH supported by %s.", + sc->sc_dconsumer->provider->name); + } else { + GJ_DEBUG(0, "BIO_FLUSH not supported by %s.", + sc->sc_dconsumer->provider->name); + } + } + + gp = sc->sc_geom; + g_topology_lock(); + pp = g_new_providerf(gp, "%s.journal", sc->sc_name); + KASSERT(pp != NULL, ("Cannot create %s.journal.", sc->sc_name)); + pp->mediasize = sc->sc_mediasize; + /* + * There could be a problem when data provider and journal providers + * have different sectorsize, but such scenario is prevented on journal + * creation. + */ + pp->sectorsize = sc->sc_sectorsize; + g_error_provider(pp, 0); + g_topology_unlock(); + last_write = time_second; + + for (;;) { + /* Get first request from the queue. */ + mtx_lock(&sc->sc_mtx); + bp = bioq_first(&sc->sc_back_queue); + if (bp != NULL) + type = (bp->bio_cflags & GJ_BIO_MASK); + if (bp == NULL) { + bp = bioq_first(&sc->sc_regular_queue); + if (bp != NULL) + type = GJ_BIO_REGULAR; + } + if (bp == NULL) { +try_switch: + if ((sc->sc_flags & GJF_DEVICE_SWITCH) || + (sc->sc_flags & GJF_DEVICE_DESTROY)) { + if (sc->sc_current_count > 0) { + mtx_unlock(&sc->sc_mtx); + g_journal_flush(sc); + g_journal_flush_send(sc); + continue; + } + if (sc->sc_flush_in_progress > 0) + goto sleep; + if (sc->sc_copy_in_progress > 0) + goto sleep; + } + if (sc->sc_flags & GJF_DEVICE_SWITCH) { + mtx_unlock(&sc->sc_mtx); + g_journal_switch(sc); + wakeup(&sc->sc_journal_copying); + continue; + } + if (sc->sc_flags & GJF_DEVICE_DESTROY) { + GJ_DEBUG(1, "Shutting down worker " + "thread for %s.", gp->name); + sc->sc_worker = NULL; + wakeup(&sc->sc_worker); + mtx_unlock(&sc->sc_mtx); + kthread_exit(0); + } +sleep: + g_journal_wait(sc, last_write); + continue; + } + /* + * If we're in switch process, we need to delay all new + * write requests until its done. + */ + if ((sc->sc_flags & GJF_DEVICE_SWITCH) && + type == GJ_BIO_REGULAR && bp->bio_cmd == BIO_WRITE) { + GJ_LOGREQ(2, bp, "WRITE on SWITCH"); + goto try_switch; + } + if (type == GJ_BIO_REGULAR) + bioq_remove(&sc->sc_regular_queue, bp); + else + bioq_remove(&sc->sc_back_queue, bp); + mtx_unlock(&sc->sc_mtx); + switch (type) { + case GJ_BIO_REGULAR: + /* Regular request. */ + switch (bp->bio_cmd) { + case BIO_READ: + g_journal_read(sc, bp, bp->bio_offset, + bp->bio_offset + bp->bio_length); + break; + case BIO_WRITE: + last_write = time_second; + g_journal_add_request(sc, bp); + g_journal_flush_send(sc); + break; + default: + panic("Invalid bio_cmd (%d).", bp->bio_cmd); + } + break; + case GJ_BIO_COPY: + switch (bp->bio_cmd) { + case BIO_READ: + if (g_journal_copy_read_done(bp)) + g_journal_copy_send(sc); + break; + case BIO_WRITE: + g_journal_copy_write_done(bp); + g_journal_copy_send(sc); + break; + default: + panic("Invalid bio_cmd (%d).", bp->bio_cmd); + } + break; + case GJ_BIO_JOURNAL: + g_journal_flush_done(bp); + g_journal_flush_send(sc); + break; + case GJ_BIO_READ: + default: + panic("Invalid bio (%d).", type); + } + } +} + +static void +g_journal_destroy_event(void *arg, int flags __unused) +{ + struct g_journal_softc *sc; + + g_topology_assert(); + sc = arg; + g_journal_destroy(sc); +} + +static void +g_journal_timeout(void *arg) +{ + struct g_journal_softc *sc; + + sc = arg; + GJ_DEBUG(0, "Timeout. Journal %s cannot be completed.", + sc->sc_geom->name); + g_post_event(g_journal_destroy_event, sc, M_NOWAIT, NULL); +} + +static struct g_geom * +g_journal_create(struct g_class *mp, struct g_provider *pp, + const struct g_journal_metadata *md) +{ + struct g_journal_softc *sc; + struct g_geom *gp; + struct g_consumer *cp; + int error; + + g_topology_assert(); + /* + * There are two possibilities: + * 1. Data and both journals are on the same provider. + * 2. Data and journals are all on separated providers. + */ + /* Look for journal device with the same ID. */ + LIST_FOREACH(gp, &mp->geom, geom) { + sc = gp->softc; + if (sc == NULL) + continue; + if (sc->sc_id == md->md_id) + break; + } + if (gp == NULL) + sc = NULL; + else if (sc != NULL && (sc->sc_type & md->md_type) != 0) { + GJ_DEBUG(1, "Journal device %u already configured.", sc->sc_id); + return (NULL); + } + if (md->md_type == 0 || (md->md_type & ~GJ_TYPE_COMPLETE) != 0) { + GJ_DEBUG(0, "Invalid type on %s.", pp->name); + return (NULL); + } + if (md->md_type & GJ_TYPE_DATA) { + GJ_DEBUG(0, "Journal %u: %s contains data.", md->md_id, + pp->name); + } + if (md->md_type & GJ_TYPE_JOURNAL) { + GJ_DEBUG(0, "Journal %u: %s contains journal.", md->md_id, + pp->name); + } + + if (sc == NULL) { + /* Action geom. */ + sc = malloc(sizeof(*sc), M_JOURNAL, M_WAITOK | M_ZERO); + sc->sc_id = md->md_id; + sc->sc_type = 0; + sc->sc_flags = 0; + sc->sc_worker = NULL; + + gp = g_new_geomf(mp, "gjournal %u", sc->sc_id); + gp->start = g_journal_start; + gp->orphan = g_journal_orphan; + gp->access = g_journal_access; + gp->softc = sc; + sc->sc_geom = gp; + + mtx_init(&sc->sc_mtx, "gjournal", NULL, MTX_DEF); + + bioq_init(&sc->sc_back_queue); + bioq_init(&sc->sc_regular_queue); + bioq_init(&sc->sc_delayed_queue); + sc->sc_delayed_count = 0; + sc->sc_current_queue = NULL; + sc->sc_current_count = 0; + sc->sc_flush_queue = NULL; + sc->sc_flush_count = 0; + sc->sc_flush_in_progress = 0; + sc->sc_copy_queue = NULL; + sc->sc_copy_in_progress = 0; + sc->sc_inactive.jj_queue = NULL; + sc->sc_active.jj_queue = NULL; + + callout_init(&sc->sc_callout, CALLOUT_MPSAFE); + if (md->md_type != GJ_TYPE_COMPLETE) { + /* + * Journal and data are on separate providers. + * At this point we have only one of them. + * We setup a timeout in case the other part will not + * appear, so we won't wait forever. + */ + callout_reset(&sc->sc_callout, 5 * hz, + g_journal_timeout, sc); + } + } + + /* Remember type of the data provider. */ + if (md->md_type & GJ_TYPE_DATA) + sc->sc_orig_type = md->md_type; + sc->sc_type |= md->md_type; + cp = NULL; + + if (md->md_type & GJ_TYPE_DATA) { + if (md->md_flags & GJ_FLAG_CLEAN) + sc->sc_flags |= GJF_DEVICE_CLEAN; + if (md->md_flags & GJ_FLAG_CHECKSUM) + sc->sc_flags |= GJF_DEVICE_CHECKSUM; + cp = g_new_consumer(gp); + error = g_attach(cp, pp); + KASSERT(error == 0, ("Cannot attach to %s (error=%d).", + pp->name, error)); + error = g_access(cp, 1, 1, 1); + if (error != 0) { + GJ_DEBUG(0, "Cannot access %s (error=%d).", pp->name, + error); + g_journal_destroy(sc); + return (NULL); + } + sc->sc_dconsumer = cp; + sc->sc_mediasize = pp->mediasize - pp->sectorsize; + sc->sc_sectorsize = pp->sectorsize; + sc->sc_jstart = md->md_jstart; + sc->sc_jend = md->md_jend; + if (md->md_provider[0] != '\0') + sc->sc_flags |= GJF_DEVICE_HARDCODED; + sc->sc_journal_offset = md->md_joffset; + sc->sc_journal_id = md->md_jid; + sc->sc_journal_previous_id = md->md_jid; + } + if (md->md_type & GJ_TYPE_JOURNAL) { + if (cp == NULL) { + cp = g_new_consumer(gp); + error = g_attach(cp, pp); + KASSERT(error == 0, ("Cannot attach to %s (error=%d).", + pp->name, error)); + error = g_access(cp, 1, 1, 1); + if (error != 0) { + GJ_DEBUG(0, "Cannot access %s (error=%d).", + pp->name, error); + g_journal_destroy(sc); + return (NULL); + } + } else { + /* + * Journal is on the same provider as data, which means + * that data provider ends where journal starts. + */ + sc->sc_mediasize = md->md_jstart; + } + sc->sc_jconsumer = cp; + } + + if ((sc->sc_type & GJ_TYPE_COMPLETE) != GJ_TYPE_COMPLETE) { + /* Journal is not complete yet. */ + return (gp); + } else { + /* Journal complete, cancel timeout. */ + callout_drain(&sc->sc_callout); + } + + error = kthread_create(g_journal_worker, sc, &sc->sc_worker, 0, 0, + "g_journal %s", sc->sc_name); + if (error != 0) { + GJ_DEBUG(0, "Cannot create worker thread for %s.journal.", + sc->sc_name); + g_journal_destroy(sc); + return (NULL); + } + + return (gp); +} + +static void +g_journal_destroy_consumer(void *arg, int flags __unused) +{ + struct g_consumer *cp; + + g_topology_assert(); + cp = arg; + g_detach(cp); + g_destroy_consumer(cp); +} + +static int +g_journal_destroy(struct g_journal_softc *sc) +{ + struct g_geom *gp; + struct g_provider *pp; + struct g_consumer *cp; + + g_topology_assert(); + + if (sc == NULL) + return (ENXIO); + + gp = sc->sc_geom; + pp = LIST_FIRST(&gp->provider); + if (pp != NULL) { + if (pp->acr != 0 || pp->acw != 0 || pp->ace != 0) { + GJ_DEBUG(1, "Device %s is still open (r%dw%de%d).", + pp->name, pp->acr, pp->acw, pp->ace); + return (EBUSY); + } + g_error_provider(pp, ENXIO); + + g_journal_flush(sc); + g_journal_flush_send(sc); + g_journal_switch(sc); + } + + sc->sc_flags |= (GJF_DEVICE_DESTROY | GJF_DEVICE_CLEAN); + + g_topology_unlock(); + callout_drain(&sc->sc_callout); + mtx_lock(&sc->sc_mtx); + wakeup(sc); + while (sc->sc_worker != NULL) + msleep(&sc->sc_worker, &sc->sc_mtx, PRIBIO, "gj:destroy", 0); + mtx_destroy(&sc->sc_mtx); + + if (pp != NULL) { + GJ_DEBUG(1, "Marking %s as clean.", sc->sc_name); + g_journal_metadata_update(sc); + g_topology_lock(); + pp->flags |= G_PF_WITHER; + g_orphan_provider(pp, ENXIO); + } else { + g_topology_lock(); + } + + if (sc->sc_current_count != 0) { + GJ_DEBUG(0, "Warning! Number of current requests %d.", + sc->sc_current_count); + } + + LIST_FOREACH(cp, &gp->consumer, consumer) { + if (cp->acr + cp->acw + cp->ace > 0) + g_access(cp, -1, -1, -1); + /* + * We keep all consumers open for writting, so if I'll detach + * and destroy consumer here, I'll get providers for taste, so + * journal will be started again. + * Sending an event here, prevents this from happening. + */ + g_post_event(g_journal_destroy_consumer, cp, M_WAITOK, NULL); + } + gp->softc = NULL; + g_wither_geom(gp, ENXIO); + free(sc, M_JOURNAL); + return (0); +} + +static void +g_journal_taste_orphan(struct g_consumer *cp) +{ + + KASSERT(1 == 0, ("%s called while tasting %s.", __func__, + cp->provider->name)); +} + +static struct g_geom * +g_journal_taste(struct g_class *mp, struct g_provider *pp, int flags __unused) +{ + struct g_journal_metadata md; + struct g_consumer *cp; + struct g_geom *gp; + int error; + + g_topology_assert(); + g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name); + GJ_DEBUG(2, "Tasting %s.", pp->name); + if (pp->geom->class == mp) + return (NULL); + + gp = g_new_geomf(mp, "journal:taste"); + /* This orphan function should be never called. */ + gp->orphan = g_journal_taste_orphan; + cp = g_new_consumer(gp); + g_attach(cp, pp); + error = g_journal_metadata_read(cp, &md); + g_detach(cp); + g_destroy_consumer(cp); + g_destroy_geom(gp); + if (error != 0) + return (NULL); + gp = NULL; + + if (md.md_provider[0] != '\0' && strcmp(md.md_provider, pp->name) != 0) + return (NULL); + if (md.md_provsize != 0 && md.md_provsize != pp->mediasize) + return (NULL); + if (g_journal_debug >= 2) + journal_metadata_dump(&md); + + gp = g_journal_create(mp, pp, &md); + return (gp); +} + +static struct g_journal_softc * +g_journal_find_device(struct g_class *mp, const char *name) +{ + struct g_journal_softc *sc; + struct g_geom *gp; + struct g_provider *pp; + + if (strncmp(name, "/dev/", 5) == 0) + name += 5; + LIST_FOREACH(gp, &mp->geom, geom) { + sc = gp->softc; + if (sc == NULL) + continue; + if (sc->sc_flags & GJF_DEVICE_DESTROY) + continue; + if ((sc->sc_type & GJ_TYPE_COMPLETE) != GJ_TYPE_COMPLETE) + continue; + pp = LIST_FIRST(&gp->provider); + if (strcmp(sc->sc_name, name) == 0) + return (sc); + if (pp != NULL && strcmp(pp->name, name) == 0) + return (sc); + } + return (NULL); +} + +static void +g_journal_ctl_destroy(struct gctl_req *req, struct g_class *mp) +{ + struct g_journal_softc *sc; + const char *name; + char param[16]; + int *nargs; + int error, i; + + g_topology_assert(); + + nargs = gctl_get_paraml(req, "nargs", sizeof(*nargs)); + if (nargs == NULL) { + gctl_error(req, "No '%s' argument.", "nargs"); + return; + } + if (*nargs <= 0) { + gctl_error(req, "Missing device(s)."); + return; + } + + for (i = 0; i < *nargs; i++) { + snprintf(param, sizeof(param), "arg%d", i); + name = gctl_get_asciiparam(req, param); + if (name == NULL) { + gctl_error(req, "No 'arg%d' argument.", i); + return; + } + sc = g_journal_find_device(mp, name); + if (sc == NULL) { + gctl_error(req, "No such device: %s.", name); + return; + } + error = g_journal_destroy(sc); + if (error != 0) { + gctl_error(req, "Cannot destroy device %s (error=%d).", + LIST_FIRST(&sc->sc_geom->provider)->name, error); + return; + } + } +} + +static void +g_journal_config(struct gctl_req *req, struct g_class *mp, const char *verb) +{ + uint32_t *version; + + g_topology_assert(); + + version = gctl_get_paraml(req, "version", sizeof(*version)); + if (version == NULL) { + gctl_error(req, "No '%s' argument.", "version"); + return; + } + if (*version != G_JOURNAL_VERSION) { + gctl_error(req, "Userland and kernel parts are out of sync."); + return; + } + + if (strcmp(verb, "destroy") == 0 || strcmp(verb, "stop") == 0) { + g_journal_ctl_destroy(req, mp); + return; + } + + gctl_error(req, "Unknown verb."); +} + +static void +g_journal_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp, + struct g_consumer *cp, struct g_provider *pp) +{ + struct g_journal_softc *sc; + + g_topology_assert(); + + sc = gp->softc; + if (sc == NULL) + return; + if (pp != NULL) { + /* Nothing here. */ + } else if (cp != NULL) { + int first = 1; + + sbuf_printf(sb, "%s", indent); + if (cp == sc->sc_dconsumer) { + sbuf_printf(sb, "Data"); + first = 0; + } + if (cp == sc->sc_jconsumer) { + if (!first) + sbuf_printf(sb, ","); + sbuf_printf(sb, "Journal"); + } + sbuf_printf(sb, "\n"); + if (cp == sc->sc_jconsumer) { + sbuf_printf(sb, "%jd", + (intmax_t)sc->sc_jstart); + sbuf_printf(sb, "%jd", + (intmax_t)sc->sc_jend); + } + } else { + sbuf_printf(sb, "%s%u\n", indent, (u_int)sc->sc_id); + } +} + +static eventhandler_tag g_journal_event_shutdown = NULL; +static eventhandler_tag g_journal_event_lowmem = NULL; + +static void +g_journal_shutdown(void *arg, int howto __unused) +{ + struct g_class *mp; + struct g_geom *gp, *gp2; + + mp = arg; + DROP_GIANT(); + g_topology_lock(); + LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) { + if (gp->softc == NULL) + continue; + GJ_DEBUG(0, "Shutting down geom %s.", gp->name); + g_journal_destroy(gp->softc); + } + g_topology_unlock(); + PICKUP_GIANT(); +} + +/* + * Free cached requests from inactive queue in case of low memory. + * We free GJ_FREE_AT_ONCE elements at once. + */ +#define GJ_FREE_AT_ONCE 4 +static void +g_journal_lowmem(void *arg, int howto __unused) +{ + struct g_journal_softc *sc; + struct g_class *mp; + struct g_geom *gp; + struct bio *bp; + u_int nfree = GJ_FREE_AT_ONCE; + + g_journal_stats_low_mem++; + mp = arg; + DROP_GIANT(); + g_topology_lock(); + LIST_FOREACH(gp, &mp->geom, geom) { + sc = gp->softc; + if (sc == NULL || (sc->sc_flags & GJF_DEVICE_DESTROY)) + continue; + mtx_lock(&sc->sc_mtx); + for (bp = sc->sc_inactive.jj_queue; nfree > 0 && bp != NULL; + nfree--, bp = bp->bio_next) { + /* + * This is safe to free the bio_data, because: + * 1. If bio_data is NULL it will be read from the + * inactive journal. + * 2. If bp is sent down, it is first removed from the + * inactive queue, so it's impossible to free the + * data from under in-flight bio. + * On the other hand, freeing elements from the active + * queue, is not safe. + */ + if (bp->bio_data != NULL) { + GJ_DEBUG(2, "Freeing data from %s.", + sc->sc_name); + gj_free(bp->bio_data, bp->bio_length); + bp->bio_data = NULL; + } + } + mtx_unlock(&sc->sc_mtx); + if (nfree == 0) + break; + } + g_topology_unlock(); + PICKUP_GIANT(); +} + +static void g_journal_switcher(void *arg); + +static void +g_journal_init(struct g_class *mp) +{ + int error; + + /* Pick a conservative value if provided value sucks. */ + if (g_journal_cache_divisor <= 0 || + (vm_kmem_size / g_journal_cache_divisor == 0)) { + g_journal_cache_divisor = 5; + } + if (g_journal_cache_limit > 0) { + g_journal_cache_limit = vm_kmem_size / g_journal_cache_divisor; + g_journal_cache_low = + (g_journal_cache_limit / 100) * g_journal_cache_switch; + } + g_journal_event_shutdown = EVENTHANDLER_REGISTER(shutdown_post_sync, + g_journal_shutdown, mp, EVENTHANDLER_PRI_FIRST); + if (g_journal_event_shutdown == NULL) + GJ_DEBUG(0, "Warning! Cannot register shutdown event."); + g_journal_event_lowmem = EVENTHANDLER_REGISTER(vm_lowmem, + g_journal_lowmem, mp, EVENTHANDLER_PRI_FIRST); + if (g_journal_event_lowmem == NULL) + GJ_DEBUG(0, "Warning! Cannot register lowmem event."); + error = kthread_create(g_journal_switcher, mp, NULL, 0, 0, + "g_journal switcher"); + KASSERT(error == 0, ("Cannot create switcher thread.")); +} + +static void +g_journal_fini(struct g_class *mp) +{ + + if (g_journal_event_shutdown != NULL) { + EVENTHANDLER_DEREGISTER(shutdown_post_sync, + g_journal_event_shutdown); + } + if (g_journal_event_lowmem != NULL) + EVENTHANDLER_DEREGISTER(vm_lowmem, g_journal_event_lowmem); + g_journal_switcher_state = GJ_SWITCHER_DIE; + wakeup(&g_journal_switcher_state); + while (g_journal_switcher_state != GJ_SWITCHER_DIED) + tsleep(&g_journal_switcher_state, PRIBIO, "jfini:wait", hz / 5); + GJ_DEBUG(1, "Switcher died."); +} + +DECLARE_GEOM_CLASS(g_journal_class, g_journal); + +static const struct g_journal_desc * +g_journal_find_desc(const char *fstype) +{ + const struct g_journal_desc *desc; + int i; + + for (desc = g_journal_filesystems[i = 0]; desc != NULL; + desc = g_journal_filesystems[++i]) { + if (strcmp(desc->jd_fstype, fstype) == 0) + break; + } + return (desc); +} + +static void +g_journal_switch_wait(struct g_journal_softc *sc) +{ + struct bintime bt; + + mtx_assert(&sc->sc_mtx, MA_OWNED); + if (g_journal_debug >= 2) { + if (sc->sc_flush_in_progress > 0) { + GJ_DEBUG(2, "%d requests flushing.", + sc->sc_flush_in_progress); + } + if (sc->sc_copy_in_progress > 0) { + GJ_DEBUG(2, "%d requests copying.", + sc->sc_copy_in_progress); + } + if (sc->sc_flush_count > 0) { + GJ_DEBUG(2, "%d requests to flush.", + sc->sc_flush_count); + } + if (sc->sc_delayed_count > 0) { + GJ_DEBUG(2, "%d requests delayed.", + sc->sc_delayed_count); + } + } + g_journal_stats_switches++; + if (sc->sc_copy_in_progress > 0) + g_journal_stats_wait_for_copy++; + GJ_TIMER_START(3, &bt); + sc->sc_flags &= ~GJF_DEVICE_BEFORE_SWITCH; + sc->sc_flags |= GJF_DEVICE_SWITCH; + wakeup(sc); + while (sc->sc_flags & GJF_DEVICE_SWITCH) { + msleep(&sc->sc_journal_copying, &sc->sc_mtx, PRIBIO, + "gj:switch", 0); + } + GJ_TIMER_STOP(3, &bt, "Switch time of %s", sc->sc_name); +} + +static void +g_journal_do_switch(struct g_class *classp, struct thread *td) +{ + struct g_journal_softc *sc; + const struct g_journal_desc *desc; + struct g_geom *gp; + struct mount *mp; + struct bintime bt; + char *mountpoint; + int asyncflag, error, vfslocked; + + DROP_GIANT(); + g_topology_lock(); + LIST_FOREACH(gp, &classp->geom, geom) { + sc = gp->softc; + if (sc == NULL) + continue; + if (sc->sc_flags & GJF_DEVICE_DESTROY) + continue; + if ((sc->sc_type & GJ_TYPE_COMPLETE) != GJ_TYPE_COMPLETE) + continue; + mtx_lock(&sc->sc_mtx); + sc->sc_flags |= GJF_DEVICE_BEFORE_SWITCH; + mtx_unlock(&sc->sc_mtx); + } + g_topology_unlock(); + PICKUP_GIANT(); + + mtx_lock(&mountlist_mtx); + TAILQ_FOREACH(mp, &mountlist, mnt_list) { + if (!(mp->mnt_flag & MNT_GJOURNAL)) + continue; + desc = g_journal_find_desc(mp->mnt_stat.f_fstypename); + if (desc == NULL) + continue; + if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) + continue; + /* mtx_unlock(&mountlist_mtx) was done inside vfs_busy() */ + + DROP_GIANT(); + g_topology_lock(); + sc = g_journal_find_device(classp, mp->mnt_stat.f_mntfromname); + g_topology_unlock(); + PICKUP_GIANT(); + + if (sc == NULL) { + GJ_DEBUG(0, "Cannot find journal geom for %s.", + mp->mnt_stat.f_mntfromname); + goto next; + } else if (JEMPTY(sc)) { + mtx_lock(&sc->sc_mtx); + sc->sc_flags &= ~GJF_DEVICE_BEFORE_SWITCH; + mtx_unlock(&sc->sc_mtx); + GJ_DEBUG(3, "No need for %s switch.", sc->sc_name); + goto next; + } + + mountpoint = mp->mnt_stat.f_mntonname; + + vfslocked = VFS_LOCK_GIANT(mp); + + error = vn_start_write(NULL, &mp, V_WAIT); + if (error != 0) { + VFS_UNLOCK_GIANT(vfslocked); + GJ_DEBUG(0, "vn_start_write(%s) failed (error=%d).", + mountpoint, error); + goto next; + } + asyncflag = mp->mnt_flag & MNT_ASYNC; + mp->mnt_flag &= ~MNT_ASYNC; + + GJ_TIMER_START(3, &bt); + vfs_msync(mp, MNT_NOWAIT); + GJ_TIMER_STOP(3, &bt, "Msync time of %s", mountpoint); + + GJ_TIMER_START(3, &bt); + error = VFS_SYNC(mp, MNT_NOWAIT, curthread); + if (error == 0) + GJ_TIMER_STOP(3, &bt, "Sync time of %s", mountpoint); + else { + GJ_DEBUG(0, "Cannot sync file system %s (error=%d).", + mountpoint, error); + } + mp->mnt_flag |= asyncflag; + + vn_finished_write(mp); + + if (error != 0) { + VFS_UNLOCK_GIANT(vfslocked); + goto next; + } + + GJ_TIMER_START(3, &bt); + error = vfs_write_suspend(mp); + VFS_UNLOCK_GIANT(vfslocked); + GJ_TIMER_STOP(3, &bt, "Suspend time of %s", mountpoint); + if (error != 0) { + GJ_DEBUG(0, "Cannot suspend file system %s (error=%d).", + mountpoint, error); + goto next; + } + + error = desc->jd_clean(mp); + if (error != 0) + goto next; + + mtx_lock(&sc->sc_mtx); + g_journal_switch_wait(sc); + mtx_unlock(&sc->sc_mtx); + + vfs_write_resume(mp); +next: + mtx_lock(&mountlist_mtx); + vfs_unbusy(mp, td); + } + mtx_unlock(&mountlist_mtx); + + sc = NULL; + for (;;) { + DROP_GIANT(); + g_topology_lock(); + LIST_FOREACH(gp, &g_journal_class.geom, geom) { + sc = gp->softc; + if (sc == NULL) + continue; + mtx_lock(&sc->sc_mtx); + if ((sc->sc_type & GJ_TYPE_COMPLETE) == GJ_TYPE_COMPLETE && + !(sc->sc_flags & GJF_DEVICE_DESTROY) && + (sc->sc_flags & GJF_DEVICE_BEFORE_SWITCH)) { + break; + } + mtx_unlock(&sc->sc_mtx); + sc = NULL; + } + g_topology_unlock(); + PICKUP_GIANT(); + if (sc == NULL) + break; + mtx_assert(&sc->sc_mtx, MA_OWNED); + g_journal_switch_wait(sc); + mtx_unlock(&sc->sc_mtx); + } +} + +/* + * TODO: Switcher thread should be started on first geom creation and killed on + * last geom destruction. + */ +static void +g_journal_switcher(void *arg) +{ + struct thread *td = curthread; + struct g_class *mp; + struct bintime bt; + int error; + + mp = arg; + for (;;) { + g_journal_switcher_wokenup = 0; + error = tsleep(&g_journal_switcher_state, PRIBIO, "jsw:wait", + g_journal_switch_time * hz); + if (g_journal_switcher_state == GJ_SWITCHER_DIE) { + g_journal_switcher_state = GJ_SWITCHER_DIED; + GJ_DEBUG(1, "Switcher exiting."); + wakeup(&g_journal_switcher_state); + kthread_exit(0); + } + if (error == 0) { + GJ_DEBUG(1, "Out of cache, force switch (used=%u " + "limit=%u).", g_journal_cache_used, + g_journal_cache_limit); + } + GJ_TIMER_START(3, &bt); + g_journal_do_switch(mp, td); + GJ_TIMER_STOP(3, &bt, "Entire switch time"); + } +} --- /dev/null Thu Jun 8 16:52:32 2006 +++ sys/geom/journal/g_journal.h Thu Jun 8 16:52:49 2006 @@ -0,0 +1,379 @@ +/*- + * Copyright (c) 2005-2006 Pawel Jakub Dawidek + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * $FreeBSD$ + */ + +#ifndef _G_JOURNAL_H_ +#define _G_JOURNAL_H_ + +#include +#include +#ifdef _KERNEL +#include +#endif + +#define G_JOURNAL_CLASS_NAME "JOURNAL" + +#define G_JOURNAL_MAGIC "GEOM::JOURNAL" +/* + * Version history: + * 0 - Initial version number. + */ +#define G_JOURNAL_VERSION 0 + +#ifdef _KERNEL +extern int g_journal_debug; + +#define GJ_DEBUG(lvl, ...) do { \ + if (g_journal_debug >= (lvl)) { \ + printf("GEOM_JOURNAL"); \ + if (g_journal_debug > 0) \ + printf("[%u]", lvl); \ + printf(": "); \ + printf(__VA_ARGS__); \ + printf("\n"); \ + } \ +} while (0) +#define GJ_LOGREQ(lvl, bp, ...) do { \ + if (g_journal_debug >= (lvl)) { \ + printf("GEOM_JOURNAL"); \ + if (g_journal_debug > 0) \ + printf("[%u]", lvl); \ + printf(": "); \ + printf(__VA_ARGS__); \ + printf(" "); \ + g_print_bio(bp); \ + printf("\n"); \ + } \ +} while (0) + +#define JEMPTY(sc) ((sc)->sc_journal_offset - \ + (sc)->sc_jprovider->sectorsize == \ + (sc)->sc_active.jj_offset && \ + (sc)->sc_current_count == 0) + +#define GJ_BIO_REGULAR 0x00 +#define GJ_BIO_READ 0x01 +#define GJ_BIO_JOURNAL 0x02 +#define GJ_BIO_COPY 0x03 +#define GJ_BIO_MASK 0x0f + +#if 0 +#define GJF_BIO_DONT_FREE 0x10 +#define GJF_BIO_MASK 0xf0 +#endif + +#define GJF_DEVICE_HARDCODED 0x0001 +#define GJF_DEVICE_DESTROY 0x0010 +#define GJF_DEVICE_SWITCH 0x0020 +#define GJF_DEVICE_BEFORE_SWITCH 0x0040 +#define GJF_DEVICE_CLEAN 0x0080 +#define GJF_DEVICE_CHECKSUM 0x0100 + +#define GJ_HARD_LIMIT 64 + +/* + * We keep pointers to journaled data in bio structure and because we + * need to store two off_t values (offset in data provider and offset in + * journal), we have to borrow bio_completed field for this. + */ +#define bio_joffset bio_completed +/* + * Use bio_caller1 field as a pointer in queue. + */ +#define bio_next bio_caller1 + +/* + * There are two such structures maintained inside each journaled device. + * One describes active part of the journal, were recent requests are stored. + * The second describes the last consistent part of the journal with requests + * that are copied to the destination provider. + */ +struct g_journal_journal { + struct bio *jj_queue; /* Cached journal entries. */ + off_t jj_offset; /* Journal's start offset. */ +}; + +struct g_journal_softc { + uint32_t sc_id; + uint8_t sc_type; + uint8_t sc_orig_type; + struct g_geom *sc_geom; + u_int sc_flags; + struct mtx sc_mtx; + off_t sc_mediasize; + u_int sc_sectorsize; +#define GJ_FLUSH_DATA 0x01 +#define GJ_FLUSH_JOURNAL 0x02 + u_int sc_bio_flush; + + uint32_t sc_journal_id; + uint32_t sc_journal_next_id; + int sc_journal_copying; + off_t sc_journal_offset; + off_t sc_journal_previous_id; + + struct bio_queue_head sc_back_queue; + struct bio_queue_head sc_regular_queue; + + struct bio_queue_head sc_delayed_queue; + int sc_delayed_count; + + struct bio *sc_current_queue; + int sc_current_count; + + struct bio *sc_flush_queue; + int sc_flush_count; + int sc_flush_in_progress; + + struct bio *sc_copy_queue; + int sc_copy_in_progress; + + struct g_consumer *sc_dconsumer; + struct g_consumer *sc_jconsumer; + + struct g_journal_journal sc_inactive; + struct g_journal_journal sc_active; + + off_t sc_jstart; /* Journal space start offset. */ + off_t sc_jend; /* Journal space end offset. */ + + struct callout sc_callout; + struct proc *sc_worker; +}; +#define sc_dprovider sc_dconsumer->provider +#define sc_jprovider sc_jconsumer->provider +#define sc_name sc_dprovider->name + +#define GJQ_INSERT_HEAD(head, bp) do { \ + (bp)->bio_next = (head); \ + (head) = (bp); \ +} while (0) +#define GJQ_INSERT_AFTER(head, bp, pbp) do { \ + if ((pbp) == NULL) \ + GJQ_INSERT_HEAD(head, bp); \ + else { \ + (bp)->bio_next = (pbp)->bio_next; \ + (pbp)->bio_next = (bp); \ + } \ +} while (0) +#define GJQ_FIRST(head) (head) +#define GJQ_REMOVE(head, bp) do { \ + struct bio *_bp; \ + \ + if ((head) == (bp)) { \ + (head) = (bp)->bio_next; \ + (bp)->bio_next = NULL; \ + break; \ + } \ + for (_bp = (head); _bp->bio_next != NULL; _bp = _bp->bio_next) {\ + if (_bp->bio_next == (bp)) \ + break; \ + } \ + KASSERT(_bp->bio_next != NULL, ("NULL bio_next")); \ + KASSERT(_bp->bio_next == (bp), ("bio_next != bp")); \ + _bp->bio_next = (bp)->bio_next; \ + (bp)->bio_next = NULL; \ +} while (0) +#define GJQ_FOREACH(head, bp) \ + for ((bp) = (head); (bp) != NULL; (bp) = (bp)->bio_next) + +#define GJ_HEADER_MAGIC "GJHDR" + +struct g_journal_header { + char jh_magic[sizeof(GJ_HEADER_MAGIC)]; + uint32_t jh_journal_id; + uint32_t jh_journal_next_id; +} __packed; + +struct g_journal_entry { + uint64_t je_joffset; + uint64_t je_offset; + uint64_t je_length; +} __packed; + +#define GJ_RECORD_HEADER_MAGIC "GJRHDR" +#define GJ_RECORD_HEADER_NENTRIES (20) +#define GJ_RECORD_MAX_SIZE(sc) \ + ((sc)->sc_jprovider->sectorsize + GJ_RECORD_HEADER_NENTRIES * MAXPHYS) +#define GJ_VALIDATE_OFFSET(offset, sc) do { \ + if ((offset) + GJ_RECORD_MAX_SIZE(sc) >= (sc)->sc_jend) { \ + (offset) = (sc)->sc_jstart; \ + GJ_DEBUG(2, "Starting from the begining (%s).", \ + (sc)->sc_name); \ + } \ +} while (0) + +struct g_journal_record_header { + char jrh_magic[sizeof(GJ_RECORD_HEADER_MAGIC)]; + uint32_t jrh_journal_id; + uint16_t jrh_nentries; + u_char jrh_sum[8]; + struct g_journal_entry jrh_entries[GJ_RECORD_HEADER_NENTRIES]; +} __packed; + +typedef int (g_journal_clean_t)(struct mount *mp); +typedef void (g_journal_dirty_t)(struct g_consumer *cp); + +struct g_journal_desc { + const char *jd_fstype; + g_journal_clean_t *jd_clean; + g_journal_dirty_t *jd_dirty; +}; + +/* Supported file systems. */ +extern const struct g_journal_desc g_journal_ufs; + +#define GJ_TIMER_START(lvl, bt) do { \ + if (g_journal_debug >= (lvl)) \ + binuptime(bt); \ +} while (0) +#define GJ_TIMER_STOP(lvl, bt, ...) do { \ + if (g_journal_debug >= (lvl)) { \ + struct bintime _bt2; \ + struct timeval _tv; \ + \ + binuptime(&_bt2); \ + bintime_sub(&_bt2, bt); \ + bintime2timeval(&_bt2, &_tv); \ + printf("GEOM_JOURNAL"); \ + if (g_journal_debug > 0) \ + printf("[%u]", lvl); \ + printf(": "); \ + printf(__VA_ARGS__); \ + printf(": %jd.%06jds\n", (intmax_t)_tv.tv_sec, \ + (intmax_t)_tv.tv_usec); \ + } \ +} while (0) +#endif /* _KERNEL */ + +#define GJ_TYPE_DATA 0x01 +#define GJ_TYPE_JOURNAL 0x02 +#define GJ_TYPE_COMPLETE (GJ_TYPE_DATA|GJ_TYPE_JOURNAL) + +#define GJ_FLAG_CLEAN 0x01 +#define GJ_FLAG_CHECKSUM 0x02 + +struct g_journal_metadata { + char md_magic[16]; /* Magic value. */ + uint32_t md_version; /* Version number. */ + uint32_t md_id; /* Journal unique ID. */ + uint8_t md_type; /* Provider type. */ + uint64_t md_jstart; /* Journal space start offset. */ + uint64_t md_jend; /* Journal space end offset. */ + uint64_t md_joffset; /* Last known consistent journal offset. */ + uint32_t md_jid; /* Last known consistent journal ID. */ + uint64_t md_flags; /* Journal flags. */ + char md_provider[16]; /* Hardcoded provider. */ + uint64_t md_provsize; /* Provider's size. */ + u_char md_hash[16]; /* MD5 hash. */ +}; +static __inline void +journal_metadata_encode(struct g_journal_metadata *md, u_char *data) +{ + MD5_CTX ctx; + + bcopy(md->md_magic, data, 16); + le32enc(data + 16, md->md_version); + le32enc(data + 20, md->md_id); + *(data + 24) = md->md_type; + le64enc(data + 25, md->md_jstart); + le64enc(data + 33, md->md_jend); + le64enc(data + 41, md->md_joffset); + le32enc(data + 49, md->md_jid); + le64enc(data + 53, md->md_flags); + bcopy(md->md_provider, data + 61, 16); + le64enc(data + 77, md->md_provsize); + MD5Init(&ctx); + MD5Update(&ctx, data, 85); + MD5Final(md->md_hash, &ctx); + bcopy(md->md_hash, data + 85, 16); +} +static __inline int +journal_metadata_decode_v0(const u_char *data, struct g_journal_metadata *md) +{ + MD5_CTX ctx; + + md->md_id = le32dec(data + 20); + md->md_type = *(data + 24); + md->md_jstart = le64dec(data + 25); + md->md_jend = le64dec(data + 33); + md->md_joffset = le64dec(data + 41); + md->md_jid = le32dec(data + 49); + md->md_flags = le64dec(data + 53); + bcopy(data + 61, md->md_provider, 16); + md->md_provsize = le64dec(data + 77); + MD5Init(&ctx); + MD5Update(&ctx, data, 85); + MD5Final(md->md_hash, &ctx); + if (bcmp(md->md_hash, data + 85, 16) != 0) + return (EINVAL); + return (0); +} +static __inline int +journal_metadata_decode(const u_char *data, struct g_journal_metadata *md) +{ + int error; + + bcopy(data, md->md_magic, 16); + md->md_version = le32dec(data + 16); + switch (md->md_version) { + case 0: + error = journal_metadata_decode_v0(data, md); + break; + default: + error = EINVAL; + break; + } + return (error); +} + +static __inline void +journal_metadata_dump(const struct g_journal_metadata *md) +{ + static const char hex[] = "0123456789abcdef"; + char hash[16 * 2 + 1]; + u_int i; + + printf(" magic: %s\n", md->md_magic); + printf(" version: %u\n", (u_int)md->md_version); + printf(" id: %u\n", (u_int)md->md_id); + printf(" type: %u\n", (u_int)md->md_type); + printf(" start: %ju\n", (uintmax_t)md->md_jstart); + printf(" end: %ju\n", (uintmax_t)md->md_jend); + printf(" joffset: %ju\n", (uintmax_t)md->md_joffset); + printf(" jid: %u\n", (u_int)md->md_jid); + printf(" flags: %u\n", (u_int)md->md_flags); + printf("hcprovider: %s\n", md->md_provider); + printf(" provsize: %ju\n", (uintmax_t)md->md_provsize); + bzero(hash, sizeof(hash)); + for (i = 0; i < 16; i++) { + hash[i * 2] = hex[md->md_hash[i] >> 4]; + hash[i * 2 + 1] = hex[md->md_hash[i] & 0x0f]; + } + printf(" MD5 hash: %s\n", hash); +} +#endif /* !_G_JOURNAL_H_ */ --- /dev/null Thu Jun 8 16:52:53 2006 +++ sys/geom/journal/g_journal_ufs.c Thu Jun 8 16:52:55 2006 @@ -0,0 +1,105 @@ +/*- + * Copyright (c) 2005-2006 Pawel Jakub Dawidek + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#include +__FBSDID("$FreeBSD$"); + +#include +#include +#include +#include + +#include +#include +#include +#include +#include + +#include +#include + +#include +#include + +static const int superblocks[] = SBLOCKSEARCH; + +static int +g_journal_ufs_clean(struct mount *mp) +{ + struct ufsmount *ump; + struct fs *fs; + + ump = VFSTOUFS(mp); + fs = ump->um_fs; + + fs->fs_clean = 1; + ffs_sbupdate(ump, MNT_WAIT, 1); + fs->fs_clean = 0; + + return (0); +} + +static void +g_journal_ufs_dirty(struct g_consumer *cp) +{ + struct fs *fs; + int error, i, sb; + + if (SBLOCKSIZE % cp->provider->sectorsize != 0) + return; + for (i = 0; (sb = superblocks[i]) != -1; i++) { + if (sb % cp->provider->sectorsize != 0) + continue; + fs = g_read_data(cp, sb, SBLOCKSIZE, NULL); + if (fs == NULL) + continue; + if (fs->fs_magic != FS_UFS1_MAGIC && + fs->fs_magic != FS_UFS2_MAGIC) { + g_free(fs); + continue; + } + GJ_DEBUG(0, "clean=%d flags=0x%x", fs->fs_clean, fs->fs_flags); + fs->fs_clean = 0; + fs->fs_flags = FS_NEEDSFSCK | FS_UNCLEAN; + error = g_write_data(cp, sb, fs, SBLOCKSIZE); + g_free(fs); + if (error != 0) { + GJ_DEBUG(0, "Cannot mark file system %s as dirty " + "(error=%d).", cp->provider->name, error); + } else { + GJ_DEBUG(0, "File system %s marked as dirty.", + cp->provider->name); + } + } +} + +const struct g_journal_desc g_journal_ufs = { + .jd_fstype = "ufs", + .jd_clean = g_journal_ufs_clean, + .jd_dirty = g_journal_ufs_dirty +}; + +MODULE_DEPEND(g_journal, ufs, 1, 1, 1); --- sys/kern/vfs_lookup.c.orig +++ sys/kern/vfs_lookup.c @@ -122,7 +122,6 @@ KASSERT((cnp->cn_flags & MPSAFE) != 0 || mtx_owned(&Giant) != 0, ("NOT MPSAFE and Giant not held")); - ndp->ni_cnd.cn_cred = ndp->ni_cnd.cn_thread->td_ucred; KASSERT(cnp->cn_cred && p, ("namei: bad cred/proc")); KASSERT((cnp->cn_nameiop & (~OPMASK)) == 0, ("namei: nameiop contaminated with flags")); @@ -173,8 +172,10 @@ * Get starting point for the translation. */ FILEDESC_LOCK(fdp); - ndp->ni_rootdir = fdp->fd_rdir; - ndp->ni_topdir = fdp->fd_jdir; + if (ndp->ni_rootdir == NULL) + ndp->ni_rootdir = fdp->fd_rdir; + if (ndp->ni_topdir == NULL) + ndp->ni_topdir = fdp->fd_jdir; dp = fdp->fd_cdir; vfslocked = VFS_LOCK_GIANT(dp->v_mount); --- sys/kern/vfs_subr.c.orig +++ sys/kern/vfs_subr.c @@ -2530,6 +2530,8 @@ strcat(buf, "|VV_TEXT"); if (vp->v_vflag & VV_SYSTEM) strcat(buf, "|VV_SYSTEM"); + if (vp->v_vflag & VV_DELETED) + strcat(buf, "|VV_DELETED"); if (vp->v_iflag & VI_DOOMED) strcat(buf, "|VI_DOOMED"); if (vp->v_iflag & VI_FREE) --- sys/kern/vfs_syscalls.c.orig +++ sys/kern/vfs_syscalls.c @@ -1656,7 +1656,8 @@ restart: bwillwrite(); - NDINIT(&nd, DELETE, LOCKPARENT | LOCKLEAF | MPSAFE, pathseg, path, td); + NDINIT(&nd, DELETE, LOCKPARENT | LOCKLEAF | SAVESTART | MPSAFE, pathseg, + path, td); if ((error = namei(&nd)) != 0) return (error == EINVAL ? EPERM : error); vfslocked = NDHASGIANT(&nd); @@ -1676,6 +1677,7 @@ if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); + vrele(nd.ni_startdir); if (vp == nd.ni_dvp) vrele(vp); else @@ -1701,6 +1703,7 @@ } NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); + vrele(nd.ni_startdir); if (vp == nd.ni_dvp) vrele(vp); else @@ -3423,7 +3426,8 @@ restart: bwillwrite(); - NDINIT(&nd, DELETE, LOCKPARENT | LOCKLEAF | MPSAFE, pathseg, path, td); + NDINIT(&nd, DELETE, LOCKPARENT | LOCKLEAF | SAVESTART | MPSAFE, pathseg, + path, td); if ((error = namei(&nd)) != 0) return (error); vfslocked = NDHASGIANT(&nd); @@ -3455,6 +3459,7 @@ if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); vput(vp); + vrele(nd.ni_startdir); if (nd.ni_dvp == vp) vrele(nd.ni_dvp); else @@ -3471,6 +3476,7 @@ out: NDFREE(&nd, NDF_ONLY_PNBUF); vput(vp); + vrele(nd.ni_startdir); if (nd.ni_dvp == vp) vrele(nd.ni_dvp); else --- sys/kern/vfs_vnops.c.orig +++ sys/kern/vfs_vnops.c @@ -272,18 +272,20 @@ struct ucred *file_cred; struct thread *td; { - struct mount *mp; + struct mount *mp = NULL; int error; VFS_ASSERT_GIANT(vp->v_mount); - vn_start_write(vp, &mp, V_WAIT); + if (vp->v_mount == NULL || !(vp->v_mount->mnt_flag & MNT_GJOURNAL)) + vn_start_write(vp, &mp, V_WAIT); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); if (flags & FWRITE) vp->v_writecount--; error = VOP_CLOSE(vp, flags, file_cred, td); vput(vp); - vn_finished_write(mp); + if (mp != NULL) + vn_finished_write(mp); return (error); } --- sys/modules/geom/Makefile.orig +++ sys/modules/geom/Makefile @@ -9,6 +9,7 @@ geom_fox \ geom_gate \ geom_gpt \ + geom_journal \ geom_label \ geom_mbr \ geom_mirror \ --- /dev/null Thu Jun 8 16:52:53 2006 +++ sys/modules/geom/geom_journal/Makefile Thu Jun 8 16:52:58 2006 @@ -0,0 +1,10 @@ +# $FreeBSD$ + +.PATH: ${.CURDIR}/../../../geom/journal + +KMOD= geom_journal +SRCS= g_journal.c +SRCS+= g_journal_ufs.c +SRCS+= vnode_if.h + +.include --- sys/modules/ufs/Makefile.orig +++ sys/modules/ufs/Makefile @@ -6,7 +6,7 @@ SRCS= opt_ddb.h opt_directio.h opt_ffs.h opt_ffs_broken_fixme.h opt_mac.h \ opt_quota.h opt_suiddir.h opt_ufs.h \ vnode_if.h ufs_acl.c ufs_bmap.c ufs_dirhash.c ufs_extattr.c \ - ufs_inode.c ufs_lookup.c ufs_quota.c ufs_vfsops.c \ + ufs_gjournal.c ufs_inode.c ufs_lookup.c ufs_quota.c ufs_vfsops.c \ ufs_vnops.c ffs_alloc.c ffs_balloc.c ffs_inode.c ffs_snapshot.c \ ffs_softdep.c ffs_subr.c ffs_tables.c ffs_vfsops.c ffs_vnops.c --- sys/sys/bio.h.orig +++ sys/sys/bio.h @@ -88,6 +88,7 @@ #define BIO_WRITE 0x02 #define BIO_DELETE 0x04 #define BIO_GETATTR 0x08 +#define BIO_FLUSH 0x10 #define BIO_CMD0 0x20 /* Available for local hacks */ #define BIO_CMD1 0x40 /* Available for local hacks */ #define BIO_CMD2 0x80 /* Available for local hacks */ --- sys/sys/mount.h.orig +++ sys/sys/mount.h @@ -219,7 +219,7 @@ #define MNT_SUIDDIR 0x00100000 /* special handling of SUID on dirs */ #define MNT_SOFTDEP 0x00200000 /* soft updates being done */ #define MNT_NOSYMFOLLOW 0x00400000 /* do not follow symlinks */ -#define MNT_JAILDEVFS 0x02000000 /* jail-friendly DEVFS behaviour */ +#define MNT_GJOURNAL 0x02000000 /* GEOM journal support enabled */ #define MNT_MULTILABEL 0x04000000 /* MAC support for individual objects */ #define MNT_ACLS 0x08000000 /* ACL support enabled */ #define MNT_NOATIME 0x10000000 /* disable update of file access time */ @@ -260,15 +260,15 @@ MNT_ROOTFS | MNT_NOATIME | MNT_NOCLUSTERR| \ MNT_NOCLUSTERW | MNT_SUIDDIR | MNT_SOFTDEP | \ MNT_IGNORE | MNT_EXPUBLIC | MNT_NOSYMFOLLOW | \ - MNT_JAILDEVFS | MNT_MULTILABEL | MNT_ACLS) + MNT_MULTILABEL | MNT_ACLS | MNT_GJOURNAL) /* Mask of flags that can be updated. */ #define MNT_UPDATEMASK (MNT_NOSUID | MNT_NOEXEC | \ MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | \ MNT_NOATIME | \ - MNT_NOSYMFOLLOW | MNT_IGNORE | MNT_JAILDEVFS | \ + MNT_NOSYMFOLLOW | MNT_IGNORE | \ MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR | \ - MNT_ACLS | MNT_USER) + MNT_ACLS | MNT_USER | MNT_GJOURNAL) /* * External filesystem command modifier flags. --- sys/sys/namei.h.orig +++ sys/sys/namei.h @@ -34,6 +34,7 @@ #define _SYS_NAMEI_H_ #include +#include #include struct componentname { @@ -160,6 +161,9 @@ ndp->ni_segflg = segflg; ndp->ni_dirp = namep; ndp->ni_cnd.cn_thread = td; + ndp->ni_cnd.cn_cred = td->td_ucred; + ndp->ni_rootdir = NULL; + ndp->ni_topdir = NULL; } #define NDF_NO_DVP_RELE 0x00000001 --- sys/sys/vnode.h.orig +++ sys/sys/vnode.h @@ -253,6 +253,7 @@ #define VV_SYSTEM 0x0080 /* vnode being used by kernel */ #define VV_PROCDEP 0x0100 /* vnode is process dependent */ #define VV_NOKNOTE 0x0200 /* don't activate knotes on this vnode */ +#define VV_DELETED 0x0400 /* should be removed */ /* * Vnode attributes. A field value of VNOVAL represents a field whose value --- sys/ufs/ffs/ffs_extern.h.orig +++ sys/ufs/ffs/ffs_extern.h @@ -72,6 +72,7 @@ int ffs_reallocblks(struct vop_reallocblks_args *); int ffs_realloccg(struct inode *, ufs2_daddr_t, ufs2_daddr_t, ufs2_daddr_t, int, int, struct ucred *, struct buf **); +int ffs_sbupdate(struct ufsmount *, int, int); void ffs_setblock(struct fs *, u_char *, ufs1_daddr_t); int ffs_snapblkfree(struct fs *, struct vnode *, ufs2_daddr_t, long, ino_t); void ffs_snapremove(struct vnode *vp); --- sys/ufs/ffs/ffs_vfsops.c.orig +++ sys/ufs/ffs/ffs_vfsops.c @@ -53,6 +53,7 @@ #include #include +#include #include #include #include @@ -70,7 +71,6 @@ static uma_zone_t uma_inode, uma_ufs1, uma_ufs2; -static int ffs_sbupdate(struct ufsmount *, int, int); static int ffs_reload(struct mount *, struct thread *); static int ffs_mountfs(struct vnode *, struct mount *, struct thread *); static void ffs_oldfscompat_read(struct fs *, struct ufsmount *, @@ -105,6 +105,7 @@ }; VFS_SET(ufs_vfsops, ufs, 0); +MODULE_VERSION(ufs, 1); static b_strategy_t ffs_geom_strategy; static b_write_t ffs_bufwrite; @@ -117,7 +118,7 @@ }; static const char *ffs_opts[] = { "acls", "async", "atime", "clusterr", - "clusterw", "exec", "export", "force", "from", "multilabel", + "clusterw", "exec", "export", "force", "from", "gjournal", "multilabel", "snapshot", "suid", "suiddir", "symfollow", "sync", "update", "union", NULL }; @@ -728,6 +729,14 @@ "WARNING: %s: ACLs flag on fs but no ACLs support\n", fs->fs_fsmnt); #endif +#ifndef UFS_GJOURNAL + if ((mp->mnt_flag & MNT_GJOURNAL) != 0) { + printf( +"WARNING: %s: GJOURNAL flag given but no UFS_GJOURNAL support\n", + fs->fs_fsmnt); + mp->mnt_flag &= ~MNT_GJOURNAL; + } +#endif ump->um_mountp = mp; ump->um_dev = dev; ump->um_devvp = devvp; @@ -1441,7 +1450,7 @@ /* * Write a superblock and associated information back to disk. */ -static int +int ffs_sbupdate(mp, waitfor, suspended) struct ufsmount *mp; int waitfor; --- /dev/null Thu Jun 8 16:52:53 2006 +++ sys/ufs/ufs/gjournal.h Thu Jun 8 16:53:03 2006 @@ -0,0 +1,38 @@ +/*- + * Copyright (c) 2005-2006 Pawel Jakub Dawidek + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * $FreeBSD$ + */ + +#ifndef _UFS_UFS_GJOURNAL_H_ +#define _UFS_UFS_GJOURNAL_H_ + +/* + * GEOM journal function prototypes. + */ +int g_journal_move(struct vnode *fdvp, struct vnode *fvp, + struct componentname *fcnp); +void g_journal_remove(struct vnode *vp); +#endif /* !_UFS_UFS_GJOURNAL_H_ */ --- /dev/null Thu Jun 8 16:52:53 2006 +++ sys/ufs/ufs/ufs_gjournal.c Thu Jun 8 16:53:07 2006 @@ -0,0 +1,252 @@ +/*- + * Copyright (c) 2005-2006 Pawel Jakub Dawidek + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#include +__FBSDID("$FreeBSD$"); + +#include "opt_ufs.h" + +#ifdef UFS_GJOURNAL + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include +#include +#include +#include +#include + +struct gj_garbage { + char gc_path[PATH_MAX]; + int gc_type; + TAILQ_ENTRY(gj_garbage) gc_next; +}; +static TAILQ_HEAD(, gj_garbage) gjgc_queue = TAILQ_HEAD_INITIALIZER(gjgc_queue); +static struct mtx gjgc_mtx; +MTX_SYSINIT(gjgc, &gjgc_mtx, "gjournal garbage collector", MTX_DEF); + +int +g_journal_move(struct vnode *fdvp, struct vnode *fvp, + struct componentname *fcnp) +{ + char path[PATH_MAX]; + struct thread *td; + struct ucred *scred; + struct mount *mp; + struct vnode *tdvp, *tvp; + struct inode *ip; + struct nameidata tond; + int error, vfslocked; + + td = curthread; + scred = thread0.td_ucred; + + mp = fvp->v_mount; + if (!(mp->mnt_flag & MNT_GJOURNAL)) + return (0); + VI_LOCK(fvp); + if (fvp->v_usecount < 2 || (fvp->v_vflag & VV_DELETED)) { + VI_UNLOCK(fvp); + return (0); + } + ip = VTOI(fvp); + if ((fvp->v_type == VDIR && ip->i_nlink > 2) || + (fvp->v_type != VDIR && ip->i_nlink > 1)) { + VI_UNLOCK(fvp); + return (0); + } + fvp->v_vflag |= VV_DELETED; + VI_UNLOCK(fvp); + + snprintf(path, sizeof(path), "%s/.deleted/#%u", + mp->mnt_stat.f_mntonname, ip->i_number); +#if 0 + printf("ufs_gjournal: Move to %s.\n", path); +#endif + + VOP_UNLOCK(fdvp, 0, td); + VOP_UNLOCK(fvp, 0, td); + + vfslocked = 0; + NDINIT(&tond, RENAME, LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART | + MPSAFE, UIO_SYSSPACE, path, td); + if (fvp->v_type == VDIR) + tond.ni_cnd.cn_flags |= WILLBEDIR; + tond.ni_cnd.cn_cred = scred; + tond.ni_rootdir = rootvnode; + tond.ni_topdir = rootvnode; + if ((error = namei(&tond)) != 0) { + printf("UFS_GJGC(%u): error=%d\n", __LINE__, error); + vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY, td); + vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY, td); + return (0); + } + vfslocked = NDHASGIANT(&tond); + tdvp = tond.ni_dvp; + tvp = tond.ni_vp; + if (tvp != NULL) { + if (fvp->v_type == VDIR && tvp->v_type != VDIR) { + error = ENOTDIR; + printf("UFS_GJGC(%u): error=%d\n", __LINE__, error); + goto bad; + } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { + error = EISDIR; + printf("UFS_GJGC(%u): error=%d\n", __LINE__, error); + goto bad; + } + } + if (fvp == tdvp) { + error = EINVAL; + printf("UFS_GJGC(%u): error=%d\n", __LINE__, error); + goto bad; + } + + if (error == 0) { + vref(fdvp); + vref(fvp); + VOP_LEASE(tdvp, td, scred, LEASE_WRITE); + if (fdvp != tdvp) + VOP_LEASE(fdvp, td, scred, LEASE_WRITE); + if (tvp != NULL) + VOP_LEASE(tvp, td, scred, LEASE_WRITE); + error = VOP_RENAME(fdvp, fvp, fcnp, tdvp, tvp, &tond.ni_cnd); + if (error != 0) + printf("UFS_GJGC(%u): error=%d\n", __LINE__, error); + NDFREE(&tond, NDF_ONLY_PNBUF); + } else { +bad: + NDFREE(&tond, NDF_ONLY_PNBUF); + if (tvp != NULL) + vput(tvp); + if (tdvp == tvp) + vrele(tdvp); + else + vput(tdvp); + } + vrele(tond.ni_startdir); + vn_lock(fvp, LK_EXCLUSIVE | LK_RETRY, td); + vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY, td); + VFS_UNLOCK_GIANT(vfslocked); + if (error != 0) { + VI_LOCK(fvp); + fvp->v_vflag &= ~VV_DELETED; + VI_UNLOCK(fvp); + return (0); + } + return (1); +} + +void +g_journal_remove(struct vnode *vp) +{ + struct gj_garbage *elem; + struct mount *mp; + struct inode *ip; + + mp = vp->v_mount; + if (!(mp->mnt_flag & MNT_GJOURNAL)) + return; + if (!(vp->v_vflag & VV_DELETED)) + return; + ip = VTOI(vp); + if ((vp->v_type == VDIR && ip->i_nlink != 2) || + (vp->v_type != VDIR && ip->i_nlink != 1)) { + return; + } + + elem = malloc(sizeof(*elem), M_TEMP, M_WAITOK); + snprintf(elem->gc_path, sizeof(elem->gc_path), "%s/.deleted/#%u", + mp->mnt_stat.f_mntonname, ip->i_number); + elem->gc_type = vp->v_type; + mtx_lock(&gjgc_mtx); + TAILQ_INSERT_TAIL(&gjgc_queue, elem, gc_next); + wakeup(&gjgc_queue); + mtx_unlock(&gjgc_mtx); +} + +static struct proc *g_journal_garbage_collector_proc; + +static void +g_journal_garbage_collector(void) +{ + struct thread *td; + struct filedesc *fdp; + struct gj_garbage *elem; + int error; + + td = curthread; + fdp = td->td_proc->p_fd; + for (;;) { + mtx_lock(&gjgc_mtx); + elem = TAILQ_FIRST(&gjgc_queue); + if (elem == NULL) { + msleep(&gjgc_queue, &gjgc_mtx, PRIBIO | PDROP, + "ufsgjgc", 0); + continue; + } + TAILQ_REMOVE(&gjgc_queue, elem, gc_next); + mtx_unlock(&gjgc_mtx); + if (fdp->fd_rdir == NULL) + fdp->fd_rdir = rootvnode; + if (fdp->fd_cdir == NULL) + fdp->fd_cdir = rootvnode; + if (elem->gc_type == VDIR) + error = kern_rmdir(td, elem->gc_path, UIO_SYSSPACE); + else + error = kern_unlink(td, elem->gc_path, UIO_SYSSPACE); + if (error != 0) { + printf("UFS_GJGC: Cannot delete %s (error=%d)\n", + elem->gc_path, error); + } +#if 0 + else + printf("UFS_GJGC: %s deleted\n", elem->gc_path); +#endif + free(elem, M_TEMP); + } +} + +static struct kproc_desc ufs_gjgc_kp = { + "ufs_gjgc", + g_journal_garbage_collector, + &g_journal_garbage_collector_proc, +}; +SYSINIT(ufs_gjgc, SI_SUB_MOUNT_ROOT, SI_ORDER_ANY, kproc_start, &ufs_gjgc_kp); + +#endif /* UFS_GJOURNAL */ --- sys/ufs/ufs/ufs_inode.c.orig +++ sys/ufs/ufs/ufs_inode.c @@ -57,6 +57,9 @@ #include #include #endif +#ifdef UFS_GJOURNAL +#include +#endif /* * Last reference to an inode. If necessary, write or delete it. @@ -83,6 +86,13 @@ */ if (ip->i_mode == 0) goto out; +#ifdef UFS_GJOURNAL + /* + * For journaled file system remove file/directory from .deleted/ + * directory, which was moved there, because was open on unlink/rmdir. + */ + g_journal_remove(vp); +#endif if ((ip->i_effnlink == 0 && DOINGSOFTDEP(vp)) || (ip->i_nlink <= 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) == 0)) { --- sys/ufs/ufs/ufs_lookup.c.orig +++ sys/ufs/ufs/ufs_lookup.c @@ -388,7 +388,8 @@ (nameiop == DELETE && (ap->a_cnp->cn_flags & DOWHITEOUT) && (ap->a_cnp->cn_flags & ISWHITEOUT))) && - (flags & ISLASTCN) && dp->i_effnlink != 0) { + (flags & ISLASTCN) && dp->i_effnlink != 0 && + !(vdp->v_vflag & VV_DELETED)) { /* * Access for write is interpreted as allowing * creation of files in the directory. --- sys/ufs/ufs/ufs_vnops.c.orig +++ sys/ufs/ufs/ufs_vnops.c @@ -81,6 +81,9 @@ #ifdef UFS_DIRHASH #include #endif +#ifdef UFS_GJOURNAL +#include +#endif #include @@ -754,7 +757,7 @@ struct inode *ip; struct vnode *vp = ap->a_vp; struct vnode *dvp = ap->a_dvp; - int error; + int error = 0; struct thread *td; td = curthread; @@ -764,7 +767,15 @@ error = EPERM; goto out; } - error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0); +#ifdef UFS_GJOURNAL + /* + * For journaled file system don't remove file which is still + * open, instead, move it to .deleted/ directory until it is + * last closed. + */ + if (!g_journal_move(dvp, vp, ap->a_cnp)) +#endif + error = ufs_dirremove(dvp, ip, ap->a_cnp->cn_flags, 0); if (ip->i_nlink <= 0) vp->v_vflag |= VV_NOSYNC; if ((ip->i_flags & SF_SNAPSHOT) != 0) { @@ -1670,6 +1681,16 @@ error = EINVAL; goto out; } +#ifdef UFS_GJOURNAL + /* + * For journaled file system don't remove directory which is still open, + * instead, move it to .deleted/ directory until it is last closed. + */ + if (g_journal_move(dvp, vp, cnp)) { + error = 0; + goto out; + } +#endif /* * Delete reference to directory before purging * inode. If we crash in between, the directory